This action might not be possible to undo. Are you sure you want to continue?

**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

_

_

mπ

a

_

2

+

_

nπ

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) =

_

Ω

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

_

Ω

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

_

Ω

(U V )

,i

dΩ =

_

Γ

(U V )n

i

dΓ (2.21)

The left hand side expands to give

_

Ω

[U V

,i

+ U

,i

V ] dΩ =

_

Γ

(U V )n

i

dΓ (2.22)

which may be rearranged as

_

Ω

U V

,i

dΩ = −

_

Ω

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) =

_

Ω

W(x

i

)

_

ρc

˙

T − Q

_

dΩ −

_

Ω

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) =

_

Ω

W(x

i

)

_

ρc

˙

T − Q

_

dΩ −

_

Ω

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 =

_

Ω

W

_

ρ c

˙

T − Q

_

dΩ +

_

Ω

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

Ω

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

_

Ω

(·) dΩ ≈

_

Ω

h

(·) dΩ =

N

el

e=1

_

Ω

e

(·) dΩ (2.29)

Thus our heat equation integral becomes

G ≈ G

h

=

N

el

e=1

_

Ω

e

W

_

ρc

˙

T − Q

_

dΩ −

N

el

e=1

_

Ω

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

_

Ω

e

W

_

ρc

˙

T − Q

_

dΩ +

N

el

e=1

_

Ω

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

_

Ω

e

Wρc

˙

TdΩ = W

I

M

IJ

˙

T

J

(2.45)

where

M

IJ

=

_

Ω

e

N

I

ρ c N

J

dΩ (2.46)

deﬁnes the element heat capacity matrix. Similarly, the term

_

Ω

e

W

,i

k T

,i

dΩ = W

I

K

IJ

T

J

(2.47)

where

K

IJ

=

_

Ω

e

N

I,i

k N

J,i

dΩ (2.48)

deﬁnes the element conductivity matrix. Finally,

_

Ω

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

=

_

Ω

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

=

_

Ω

e

N

I,i

k N

J,i

dΩ (2.55)

and

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS 13

F

I

=

_

Ω

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

=

_

Ω

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

(ξ

l

) k N

J,i

(ξ

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

) =

_

Ω

W

,i

k T

η

,i

dΩ −

_

Ω

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

dη

= 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

dη

=

d

dη

(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

dη

=

_

Ω

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

dη

= 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

)

dη

= 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

)

dη

= 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) =

_

Ω

T

,i

k η T

,i

dΩ −

_

Ω

T QdΩ +

_

Γ

q

T ¯ q

n

dΓ (3.13)

The integral is trivial and gives

Π(T) =

1

2

_

Ω

T

,i

kT

,i

dΩ −

_

Ω

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 =

_

Ω

W

_

ρ c

˙

T − Q

_

dΩ +

_

Ω

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

_

Ω

∂

∂T

,i

(T

,i

k T

,i

) δT

,i

dΩ −

_

Ω

∂

∂T

(T Q) δT dΩ

+

_

Γ

q

∂

∂T

(T ¯ q

n

) δT dΓ (3.25)

Performing the derivatives leads to

δΠ =

1

2

_

Ω

(k T

,i

+ T

,i

k) δT

,i

dΩ −

_

Ω

QδT dΩ +

_

Γ

q

¯ q

n

δT dΓ (3.26)

Collecting terms we have

δΠ =

_

Ω

δT

,i

k T

,i

dΩ −

_

Ω

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

_

Ω

T

D dΩ −

_

Ω

T

D

0

dΩ

+

_

Ω

σ

T

(∇

(s)

u − ) dΩ −

_

Ω

u

T

b

v

d

Ω

−

_

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 =

_

Ω

a

T

bdΩ =

_

Ω

(a

T

b)

T

dΩ =

_

Ω

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

dη

¸

¸

¸

¸

η=0

=

_

Ω

E

T

DdΩ −

_

Ω

E

T

D

0

dΩ

+

_

Ω

S

T

(∇

(s)

u − )dΩ +

_

Ω

σ

T

(∇

(s)

U − E)dΩ

−

_

Ω

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 =

_

Ω

δ

T

DdΩ −

_

Ω

δ

T

D

0

dΩ

+

_

Ω

δσ

T

(∇

(s)

u − )dΩ +

_

Ω

σ

T

(∇

(s)

δu − δ)dΩ

−

_

Ω

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,

_

Ω

σ

T

∇

(s)

UdΩ = −

_

Ω

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

dη

¸

¸

¸

¸

η=0

=

_

Ω

E

T

[D( −

0

) − σ]dΩ

+

_

Ω

S

T

(∇

(s)

u − )dΩ −

_

Ω

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

_

Ω

0T

D

0

dΩ −

1

2

_

Ω

σ

T

D

−1

σdΩ

−

_

Ω

σ

T

0

dΩ +

_

Ω

σ

T

∇

(s)

udΩ −

_

Ω

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

_

Ω

(∇

(s)

u)

T

D(∇

(s)

u)dΩ −

_

Ω

(∇

(s)

u)

T

D

0

dΩ

−

_

Ω

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

Ω

e

(6.1)

and integrals are deﬁned as

_

Ω

( · ) dΩ ≈

_

Ω

h

( · ) dΩ =

M

el

e=1

_

Ω

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

=

_

Ω

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

__

Ω

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

=

_

Ω

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

=

_

Ω

e

(∇

(s)

u)

T

σdΩ =

e

_

Ω

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

_

Ω

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) =

_

Ω

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

=

_

Ω

e

B

T

α

σdΩ =

_

Ω

e

B

T

α

(s + p m) dΩ (6.52)

which after rearrangement gives

P

α

=

_

Ω

e

B

T

α

s dΩ +

_

Ω

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

=

_

Ω

e

(B

T

dev

)

α

s dΩ +

_

Ω

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

=

_

Ω

e

B

T

α

s dΩ +

_

Ω

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

=

_

Ω

e

(B

T

dev

)

α

s dΩ +

_

Ω

e

b

T

α

p dΩ (6.85)

Using the above deﬁnitions and identities the internal force vector may be written as

P

α

=

_

Ω

e

µ(B

T

dev

)

α

D

µ

(B

dev

)

β

dΩu

β

+

_

Ω

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

=

_

Ω

e

µ(B

T

dev

)

α

D

µ

(B

dev

)

β

dΩ =

_

Ω

e

µB

T

α

D

dev

B

β

dΩ (6.88)

and

(K

vol

)

αβ

=

_

Ω

e

Kb

α

b

T

β

dΩ =

_

Ω

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

_

Ω

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

dξ

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

dξ

2

(6.94)

and for 3-dimensions

_

2

g(ξ) dξ =

_

1

−1

_

1

−1

_

1

−1

g(ξ) dξ

1

dξ

2

dξ

3

(6.95)

Using quadrature, the stress divergence is given by

P

α

=

L

l=1

B

α

(ξ

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

(ξ

l

)

T

D(ξ

l

)B

β

(ξ

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

(ξ

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

_

Ω

T

D dΩ −

_

Ω

T

D

0

dΩ

+

_

Ω

σ

T

(∇

(s)

u − ) dΩ −

_

Ω

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

_

Ω

µ

T

D

dev

dΩ −

_

Ω

µ

T

D

dev

0

dΩ

+

_

Ω

s

T

[e(u) − e] dΩ (7.5)

+

1

2

_

Ω

K θ

2

dΩ −

_

Ω

K θ θ

0

dΩ +

_

Ω

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

_

Ω

µ

T

(u) D

dev

(u) dΩ −

_

Ω

µ

T

(u) D

dev

0

dΩ

+

1

2

_

Ω

K θ

2

dΩ −

_

Ω

K θ θ

0

dΩ

+

_

Ω

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

_

Ω

e

µB

T

I

D

dev

B

J

dΩu

J

−

_

Ω

e

µB

T

I

D

dev

0

dΩ

_

+ θ

T

_

1

2

_

Ω

e

Kφ

T

φdΩθ −

_

Ω

e

Kφ

T

θ

0

dΩ

_

+ p

T

_ _

Ω

e

φ

T

b

J

dΩu

J

−

_

Ω

e

φ

T

φdΩθ

_

(7.21)

CHAPTER 7. MIXED FINITE ELEMENT METHODS 48

If we deﬁne the following matrices:

k =

_

Ω

e

Kφ

T

φdΩ (7.22)

π

0

=

_

Ω

e

Kφ

T

θ

0

dΩ (7.23)

h =

_

Ω

e

φ

T

φdΩ (7.24)

g

I

=

_

Ω

e

φ

T

b

I

dΩ (7.25)

and recall that the deviatoric stiﬀness is deﬁned as

(K

dev

)

IJ

=

_

Ω

e

B

T

I

D

dev

B

J

dΩ (7.26)

and denote the eﬀects of initial deviatoric strains as

(P

0

dev

)

I

=

_

Ω

e

µB

T

I

D

dev

0

dΩ =

_

Ω

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

dΠ

e

dη

=

_

(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

dΠ

e

dη

= (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

dΠ

e

dη

= (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

=

_

Ω

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

(ξ

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

(ξ

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

(ξ

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

_

Ω

_

(u) θ

¸

_

D

dev

d

d

T

d

vol

_ _

(u)

θ

_

dΩ

+

_

Ω

((u) s

0

+ θ p

0

) dΩ

+

_

Ω

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

_

_

Ω

e

__

B

T

I

D

dev

B

J

B

T

I

dφ

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

_

Ω

e

φ

T

_

b

J

−φ

¸

dΩ

_

ˆ u

J

ˆ

θ

_

+

_

δˆ u

T

I

δ

ˆ

θ

T

_

_

Ω

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

_

_

Ω

e

_

B

T

I

¯s

φ

T

¯ p

_

dΩ

+ δˆ p

T

_

Ω

e

φ

T

_

b

J

−φ

¸

dΩ

_

ˆ u

J

ˆ

θ

_

+

_

δˆ u

T

I

δ

ˆ

θ

T

_

_

Ω

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

_

Ω

e

φ

T

b

J

dΩ ˆ u

J

=

_

Ω

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

_

Ω

e

φ

T

¯ p dΩ =

_

Ω

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

__

Ω

B

T

I

¯s dΩ +

¯

b

T

I

_

Ω

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

_

Ω

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 =

_

Ω

e

φ

T

φdΩ (7.81)

g

I

=

_

Ω

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 =

_

Ω

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

= −

_

Ω

e

B

T

I

σdΩ (7.89)

(K

dev

)

IJ

=

_

Ω

e

B

T

I

D

dev

B

d

J

Ω (7.90)

k

I

=

_

Ω

e

B

T

I

dφdΩ (7.91)

k

vol

=

_

Ω

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

_

Ω

e

T

D dΩ −

_

Ω

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

_

Ω

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 −

_

Ω

e

_

I

dev

B

I

u

I

+

1

3

mφ(ξ) θ

_

T

D

0

dΩ

+ p

T

_ _

Ω

e

φ

T

b

J

dΩu

J

−

_

Ω

e

φ

T

φdΩθ

_

(7.96)

For symmetric D, we can deﬁne the following matrices:

(K

dev

)

IJ

=

_

Ω

e

B

T

I

I

dev

DI

dev

B

J

dΩ (7.97)

(K

co

)

J

=

1

3

_

Ω

e

φ

T

m

T

DI

dev

B

J

dΩ (7.98)

k =

1

9

_

Ω

e

φ

T

m

T

DmφdΩ (7.99)

π

0

=

1

3

_

Ω

e

φ

T

mD

0

dΩ (7.100)

h =

_

Ω

e

φ

T

φdΩ (7.101)

g

I

=

_

Ω

e

φ

T

b

I

dΩ (7.102)

and denote the eﬀects of initial deviatoric strains as

(P

0

dev

)

I

=

_

Ω

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

dΠ

e

dη

=

_

(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

dΠ

e

dη

= (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

=

_

Ω

e

B

T

I

D

dev

B

J

dΩ (7.122)

(K

co

)

J

=

_

Ω

e

φ

T

d

T

dev

B

J

dΩ (7.123)

and

k =

_

Ω

e

d

vol

φ

T

φdΩ (7.124)

The matrix for the initial strains is computed as

π

0

=

_

Ω

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

=

_

Ω

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 =

_

Ω

e

d

vol

dΩ (7.131)

which is a 1 ×1 matrix and for constant d

vol

becomes

k = k

vol

= d

vol

Ω

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 =

_

Ω

e

dΩ = Ω

e

(7.133)

and

g

I

=

_

Ω

e

b

I

dΩ (7.134)

CHAPTER 7. MIXED FINITE ELEMENT METHODS 61

which gives the modiﬁed volumetric strain-displacement equation

¯

b

I

=

1

Ω

e

g

I

=

1

Ω

e

_

Ω

e

b

I

dΩ (7.135)

The initial strain term is computed as

π

0

=

_

Ω

e

d

T

0

dΩ (7.136)

For the mixed element the internal force is computed using

P

I

=

_

Ω

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

Ω

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

_

Ω

T

D dΩ −

_

Ω

T

D

0

dΩ

+

_

Ω

σ

T

(∇

(s)

u − ) dΩ −

_

Ω

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

_

Ω

(∇

(s)

u + ˜)

T

D(∇

(s)

u + ˜) dΩ

−

_

Ω

(∇

(s)

u + ˜)

T

D

0

dΩ −

_

Ω

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

dΠ

dη

=

_

Ω

(∇

(s)

U)

T

D(∇

(s)

u + ˜ −

0

) dΩ

+

_

Ω

˜

E

T

[D(∇

(s)

u + ˜ −

0

) − σ] dΩ −

_

Ω

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

_

Ω

e

(∇

(s)

u + ˜)

T

D(∇

(s)

u + ˜) dΩ

−

_

Ω

e

(∇

(s)

u + ˜)

T

D

0

dΩ −

_

Ω

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

dΠ

e

dη

=

_

(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

=

_

Ω

e

B

T

I

DB

J

dΩ (8.21)

˜

Γ

αJ

=

_

Ω

e

ψ

T

α

DB

J

dΩ (8.22)

˜

H

αβ

=

_

Ω

e

ψ

T

α

Dψ

β

dΩ (8.23)

Q

αβ

=

_

Ω

e

φ

T

α

ψ

β

dΩ (8.24)

P

0

I

=

_

Ω

e

B

T

I

D

0

dΩ (8.25)

˜ π

0

α

=

_

Ω

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

dΠ

e

dη

=

_

(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

dΠ

e

dη

= (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

dΠ

e

dη

¸

¸

¸

¸

η=0

=

_

Ω

e

(∇

(s)

U)

T

˜ σdΩ +

_

Ω

e

˜

E

T

˜ σdΩ (8.35)

It is noted that the orthogonality condition

_

Ω

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

−

_

Ω

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

= −

_

Ω

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

_

Ω

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)

dΠ

e

dη

=

_

Ω

e

(∇

(s)

U)

T

˜ σdΩ +

_

Ω

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

−

_

Ω

e

B

T

I

˜ σdΩ (8.57)

Similarly, the residual for the enhanced modes is computed from

˜

R

α

= −

_

Ω

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)

d˜

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

=

_

Ω

e

B

I

∂ ˜ σ

∂

¸

¸

¸

¸

(i)

∂

∂u

J

dΩ =

_

Ω

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

=

_

Ω

e

ψ

T

α

˜

D

(i)

t

B

J

dΩ (8.69)

and

˜

H

(i)

αβ

=

_

Ω

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,

d˜

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

d˜

β(i,k+1)

= 0 (8.79)

where now

˜

H

(i,k)

αβ

=

_

Ω

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

d˜

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

2λ

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

(λ

∂f

∂Σ

)

−

ˆ

H

kin

λ

∂

2

f

∂Σ

2

(I +

ˆ

H

kin

λ

∂

2

f

∂Σ

2

) −

ˆ

H

kin

∂

∂λ

(λ

∂f

∂Σ

)

∂f

∂Σ

T

−

∂f

∂Σ

T

−A

_

¸

_

_

_

dσ

(j+1)

dα

(j+1)

dλ

(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

_

_

dσ

(j+1)

dα

(j+1)

dλ

(j+1)

_

_

=

_

¸

_

ˆ

C −λ

∂

2

f

∂Σ

2

∂

∂λ

(λ

∂f

∂Σ

)

−

ˆ

H

kin

λ

∂

2

f

∂Σ

2

(I +

ˆ

H

kin

λ

∂

2

f

∂Σ

2

) −

ˆ

H

kin

∂

∂λ

(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

(λ

∂f

∂Σ

)

−

ˆ

H

kin

λ

∂f

∂Σ

(I +

ˆ

H

kin

λ

∂

2

f

∂Σ

2

) −

ˆ

H

kin

∂

∂λ

(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

_

_

dσ

(j+1)

dα

(j+1)

dλ

(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

(σ

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

(σ

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

dα

dλ

_

_

=

_

_

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

dα

_

=

_

de − ndλ

−

2

3

H

kin

ndλ

_

(10.86)

The solution to this equation is

1

_

ds

dα

_

=

_

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

(σ

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) =

_

Ω

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

dΠ

cη

dη

¸

¸

¸

¸

η=0

=

_

Ω

c

Λ

T

g(u) dΩ +

_

Ω

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 =

dΠ

cη

dη

¸

¸

¸

¸

η=0

=

_

(Λ

α

)

T

(U

I

)

T

¸

_

P

λ

α

P

λ

I

_

(11.12)

where

P

λ

α

=

e

_

Ω

ce

N

λ

α

g(u) dΩ (11.13)

and

P

λ

I

=

e

_

Ω

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

_

(Λ

α

)

T

(U

I

)

T

¸

_

0 K

λ

αJ

K

λ

Iβ

K

λ

IJ

_ _

dλ

β

du

J

_

(11.15)

where

K

λ

Iβ

=

e

_

Ω

ce

N

I

G

T

N

λ

β

dΩ (11.16)

K

λ

αJ

=

e

_

Ω

ce

N

λ

α

GN

J

dΩ = (K

λ

Jα

)

T

(11.17)

and

K

λ

IJ

=

e

_

Ω

ce

N

I

λ

T

∂

2

g

∂u∂u

N

J

dΩ =

e

_

Ω

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) =

_

Ω

c

λ

T

g(u) dΩ −

_

Ω

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.

dΠ

cη

dη

¸

¸

¸

¸

η=0

=

_

Ω

c

Λ

T

(g(u) −

1

k

λ) dΩ +

_

Ω

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

_

(Λ

α

)

T

(U

I

)

T

¸

_

K

λ

αβ

K

λ

αJ

K

λ

Iβ

K

λ

IJ

_ _

dλ

β

du

J

_

(11.22)

where

K

λ

αβ

=

e

_

Ω

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

Iβ

K

IJ

+ K

λ

IJ

_ _

dλ

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

Iβ

K

IJ

+ K

λ

IJ

_ _

dλ

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

dλ

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

dλ

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

Iβ

_

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

) =

_

Ω

c

(λ + λ

A

)

T

g(u) dΩ −

_

Ω

c

1

k

λ

T

λdΩ (11.31)

where λ

A

is the augmented term. The variation to Eq. 11.31 gives

dΠ

cη

dη

¸

¸

¸

¸

η=0

=

_

Ω

c

Λ

T

_

g(u) −

1

k

λ

_

dΩ +

_

Ω

c

U

T

G

T

(λ + λ

A

) dΩ (11.32)

+

_

Ω

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

Iβ

K

IJ

+ K

λ

IJ

_ _

dλ

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

_

Ω

ce

N

I

G

T

N

λ

β

(λ

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

dΠ

η

dη

¸

¸

¸

¸

η=0

=

e

(U

I

)

T

__

Ω

e

B

T

I

˜ σdΩ −

_

Ω

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

dΠ

η

dη

¸

¸

¸

¸

η=0

=

e

(U

I

)

T

__

Ω

e

B

T

I

˜ σdΩ +

_

Ω

e

N

I

ρ ¨ udΩ −

_

Ω

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

_

Ω

e

B

T

I

˜ σdΩ −

e

_

Ω

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

_

Ω

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

_

Ω

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

,

Ω

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

_

Ω

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 =

_

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, R, to the rate of

change of the body momentum, ˙ p. Accordingly,

_

Ω

ρ b

m

dv +

_

∂Ω

t

d

s =

_

Ω

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

_

Ω

[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

_

Ω

0

ρ

0

b

m

dV +

_

∂Ω

0

t

0

dS =

_

Ω

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

_

Ω

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) =

_

Ω

0

W(C(u)) dV −

_

Ω

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 =

_

Ω

0

∂ W

∂ C

δCdV −

_

Ω

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 =

_

Ω

0

δu

T

ρ

0

˙ v dV +

_

Ω

0

δF

T

PdV (13.80)

−

_

Ω

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 =

_

Ω

δu

T

ρ ˙ v dv +

_

Ω

∇(δu)

T

σ] dv (13.83)

−

_

Ω

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.

_

Ω

0

tr (δF

T

FS) dV =

_

Ω

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

∆(

_

Ω

0

tr (δFSF

T

dV ) =

_

Ω

0

tr (δFS∆F

T

) dV +

_

Ω

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

∆(

_

Ω

0

tr (δFSF

T

dV ) =

_

Ω

tr (∇(δu) σ∇(∆u)

T

) dv

+ +

_

Ω

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

_

Ω

tr (∇(δu) ∆σ) dv =

_

Ω

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

Ω

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

_

Ω

( · ) dv ≈

_

Ω

h

( · ) dv =

e

_

Ω

e

( · ) dv (13.97)

Using this approximation the variational equation become

e

_ _

Ω

e

δu

T

ρ ˙ v dv +

_

Ω

e

∇(δu)

T

σdv

_

=

CHAPTER 13. FINITE DEFORMATION 118

e

_ _

Ω

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

_

Ω

e

δu

T

ρ ˙ v dv = (δu

T

)

I

_

Ω

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

=

_

Ω

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

_

Ω

e

tr [∇(δu) σ] dv =

_

Ω

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

_

Ω

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

_

Ω

e

δ

T

σdv = (δu

I

)

T

_

Ω

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,

_

Ω

e

tr (∇(δu) σ∇(∆u)

T

) dv = (δu

I

)

T

_

Ω

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

=

_

Ω

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

=

_

Ω

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

_

Ω

e

tr (∇(δu) ∆σ) dv =

_

Ω

e

tr (δ c I ∆) dv (13.119)

which is evaluated for a typical element. In matrix notation the right hand side becomes

_

Ω

e

tr (δ c I ∆) dv =

_

Ω

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

_

Ω

e

δ

T

D∆ dv = (δu

I

)

T

_

Ω

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

=

_

Ω

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

_

Ω

e

δu

T

ρ b

m

dv = (δu

I

)

T

_

Ω

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

+

_

Ω

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

=

_

Ω

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

+

_

Ω

e

B

T

I

σdv − f

I

_

= 0 (13.128)

which may be written in the compact matrix form

M¨ x + N(σ) = f (13.129)

CHAPTER 13. FINITE DEFORMATION 122

where N(σ) is the stress divergence vector.

Solution of this set of equations together with satisfying the material constitution and

the displacement boundary conditions, yields the solution to a problem. A common

solution procedure is to use a Newton type solution method and solve a sequence of

linear problems. Accordingly, in a Newton Method we write the momentum equation

as

R = f − M¨ x − N(σ) = 0 (13.130)

A linearization of this set of equations gives the result

M∆¨ u + K

t

∆u = R (13.131)

where

K

t

= K

m

+ K

g

(13.132)

The above description is for a standard displacement type formulation. We refer to the

method as the standard B-matrix formulation.

13.16 Mixed formulation

In the mixed formulation used, a modiﬁed deformation gradient, (as described in [19]),

is used. The mixed formulation is used to permit solution of incompressible and nearly

incompressible materials, as well as, compressible solutions which can be treated by

a standard B matrix formulations. Thus, the modiﬁed deformation gradient is based

upon a separation into volumetric and deviatoric parts as

F = F

vol

F

dev

(13.133)

where F

vol

measures volumetric part and F

dev

the deviatoric part of deformation. Since

det F measures the volumetric part we have

J = det F = det F

vol

det F

dev

(13.134)

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) =

_

Ω

0

W

_

˜

C(u, θ)

_

dV +

_

Ω

0

p (J − θ) dV (13.142)

−

_

Ω

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])

_

Ω

δu

T

ρ ˙ v dv +

_

Ω

∇δu

T

( ˜ σ

dev

+p 1

t

) dv =

_

Ω

δu

T

ρ b

m

dv +

_

∂Ω

δu

T

¯

t ds (13.144)

for the linear momentum equation,

_

Ω

δθ(

tr ˜ σ

3θ

−

p

J

) dv = 0 (13.145)

for the relationship between the mixed pressure and the trace of the stress, and

_

Ω

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

_

Ω

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 =

Ω

e

Ω

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

Ω

e0

_

Ω

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

iα

=

∂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

iα

(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

(ξ

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

(ξ

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

mε

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

mε

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 . . . . . . . . . . . .

. . . . . . . . 101 12.9 Element Technology .12Geometric stiﬀness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .CONTENTS 10. . . . . . . . . . . . . . . . . . . . . . . . . 116 13. .3 Hilber-Hughes-Taylor (HHT) Algorithm . . 118 13. . . . . . .6 Material Constitution . . . . . . 117 13. . . . . . . . . . . . . . . . . . . . 11. . . .7 Variational Description . . 114 13. . . . . . . . . . . . . . . . . 11 Augmented Lagrangian Formulations 11. . . . . 105 13.1 Adding the transient terms .8 Linearized Equations . .4 Boundary Conditions . . . . . . . .Finite Elasticity . . . . . . .2 Stress and Traction Measures . . . . . . . . .1 Constraint Equations .3 Augmented Lagrangian Method for Constraints . . . .5 Initial Conditions . . . . . . .1 Kinematics and Deformation . 100 12. . . . . . . .3 Balance of Momentum . . . . . . . .Introduction . . . . . . . . . . . . . .14Loading terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Newmark Solution of Momentum Equations . . . 12 Transient Analysis iii 85 91 93 93 94 97 100 12. . . . . . . . . . . . . . . . . . . . . . . . . .3 Isotropic plasticity: J2 Model . . . .standard B matrix formulation . . . . . . . 108 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . . . . 121 . . . . . . . . . . . . . . . . 110 13. . . . . . . . . . . . . . . . . . . . . .13Material tangent matrix . . . . . . . . . . . . . 11.2 Mixed Penalty Methods for Constraints . . . . . . 120 13. . . . . . . .11Stress Divergence Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10Consistent and Lumped Mass Matrices . . . . . . . . . . . . . . . . . . . . . . . . . 111 13. . . 119 13. . . . . 112 13. . . . . . . . . . . . . . . . . . . 104 13 Finite Deformation 105 13. .4 Isotropic viscoplasticity: J2 model . . . . . 109 13. .

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

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

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

Very few partial diﬀerential equations may be solved in closed form . in general it is not possible to treat general boundary conditions or problem shapes using this approach. 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. 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.2) which when substituted into the equation yields 3 . Linear equations are characterized by the appearance of the dependent variable(s) in linear form only.1 Strong form for problems in engineering Many problems in engineering are modeled using partial diﬀerential equations (PDE).one case being the linear wave equation in one space dimension and time. As an example consider the Poisson equation ∂2u ∂2u + = q(x. non-linear equations include nonlinear terms also. whereas. Again. 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.Chapter 2 Introduction to Strong and Weak Forms 2.1) deﬁned on the region 0 ≤ x ≤ a. 0 ≤ y ≤ b with the boundary condition u = 0 on all edges.

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

The following notation is introduced for use throughout this report. Q is the volumetric heat generation per unit volume per unit time.5) (2. The equations hold for all points xi in the domain of interest.10) (2.8) Boundary conditions are given by ¯ T (xj . ρ is density.3 Heat conduction problem: Strong form The above steps are made more concrete by considering an example. T is temperature.7) With this notation.i = and in time by ∂T ˙ T = ∂t In addition. Partial derivatives in space will be denoted by ( · ). t) = T ¯ where T is a speciﬁed temperature for points xj on the boundary. summation convention is used where d ∂( · ) ∂xi (2. 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. the divergence of the ﬂux may be written as d qi. and t is time. ΓT .i = i=1 ∂qi ∂xi (2. qi is the component of the heat ﬂux in the xi direction.4) where: d is the spatial dimension of the problem. c is speciﬁc heat.CHAPTER 2.6) ai b i = i=1 ai b i (2. Ω. and q n = qi n i = qn ¯ (2.9) . INTRODUCTION TO STRONG AND WEAK FORMS 5 2..

CHAPTER 2.11) for points in the domain. the solution process is simpliﬁed by considering weak (variational) forms.16) We note that it is necessary to compute second derivatives of the temperature to compute a solution to the diﬀerential equation. The equations are completed by giving a relationship between the gradient of temperature and the heat ﬂux (called the thermal constitutive equation). .i ).j (2.14 into Eq. 2. Initial conditions are given by ¯ T (xi .4. The Fourier law is a linear relationship given as qi = − kij T.12) where kij is a symmetric. and ni are direction ¯ cosines of the unit outward pointing normal to the boundary.i (2. Hence for an isotropic material the Fourier law becomes qi = − kT. it is only necessary to approximate ﬁrst derivatives of functions to obtain a solution. expressed as a weak form. INTRODUCTION TO STRONG AND WEAK FORMS 6 where q nn is a speciﬁed ﬂux for points xj on the ﬂux boundary.13) in which δij is the Kronecker delta function (δij = 1 for i = j. 2. we show that.14) The diﬀerential equation may be expressed in terms of temperature by substituting Eq. For an isotropic material kij = kδij (2. In the following. Ω.i + Q = ρcT (2. Γq . Thus. at time zero. 0) = T0 (xi ) (2. second rank thermal conductivity tensor. The partial diﬀerential equation together with the boundary and initial conditions is called the strong form of the problem.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. = 0 for i = j). The result is ˙ (kT.

The equation is ﬁrst written on one side of an equal sign. Applying the integration by parts to the heat equation gives G(W. which transforms the set of diﬀerential equations onto a scalar function. T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W.18) In step 3 we integrate by parts the terms involving the spatial derivatives (i.i dΩ = 0 (2. 2.i dΩ = Ω Γ (2. INTRODUCTION TO STRONG AND WEAK FORMS 7 2.CHAPTER 2.i In step 2 we integrate over the domain. qi .22) which may be rearranged as U V. we multiply Eq.20) (U V )ni dΓ (2.17) ˙ W (xi ) ρcT − Q + qi.i + U.i dΩ = − Ω Ω U.e.i V ] dΩ = Ω Γ (U V )ni dΓ (2. qi . φ is the product of two functions. G(W.4 Heat conduction problem: Weak form In step 1. T ) = Ω = 0 (2.4 by an arbitrary function W (xi ).i qi d Ω (2.19) Normally.23) which we observe is an integration by parts. the thermal ﬂux vector in our case).i V dΩ + Γ (U V )ni dΓ (2. Green’s theorem is given by φ.24) + Γ W qi ni dΓ = 0 .i dΩ = Ω Γ φ ni d Γ (2. Thus for φ = VU we have (U V ). T ) = W (xi ) ρcT − Q + qi. Thus ˙ g(W..21) The left hand side expands to give [U V. qi . Thus. Ω.

This leads to weaker conditions to deﬁne solutions of the problem and thus the notion of a weak form is established.24 completes step 4 and we obtain the ﬁnal expression G(W.CHAPTER 2. hence our equation must be valid even if W is zero for some parts of the domain). 2.10 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. thus. 2.27) + 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. qi .27 is said to be irreducible [26.i dΩ (2. 2. T ) = Ω ˙ W (xi ) ρcT − Q dΩ − Ω W. INTRODUCTION TO STRONG AND WEAK FORMS Introducing qn . Eq.i qi dΩ (2. there are no additional equations that can be used to give any additional reductions. Accordingly. the boundary term may be split into two parts and expressed as W qn dΓ = Γ ΓT 8 W qn dΓ + Γq W qn dΓ (2. 2. Substituting all the above into Eq.25) Now the boundary condition Eq.i k T. Chapter 9].5 Approximate solutions: The ﬁnite element method For ﬁnite element approximate solutions. we let Nel Ω ≈ Ωh = e=1 Ωe (2. we deﬁne each integral as a sum of integrals over each element.28) . Furthermore.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.

Commonly. This occurs under the condition that both W and T are continuous in Ω. Integrals may now be summed over each element and written as Nel (·) dΩ ≈ Ω Ωh (·) dΩ = e=1 Ωe (·) dΩ (2. which maintain the C 0 condition.29) Thus our heat equation integral becomes Nel Nel G ≈ Gh = e=1 Nel Ωe W ˙ ρcT − Q dΩ − e=1 Ωe W. Ωe is the domain of a typical element and Nel is the number of nodes attached to the element.i kT.i dΩ (2. satisfy the conditions Nel xi = I=1 NI (ξ)xI i (2. the ﬁrst derivatives of W and T may be discontinuous in Ω. surface integrals between adjacent elements must vanish. With this approximation.32) for coordinates and Nel T = I=1 NI (ξ)T I (t) (2. INTRODUCTION TO STRONG AND WEAK FORMS 9 where Ωh is the approximation to the domain created by the set of elements. but not its ﬁrst derivatives. Isoparametric elements. deﬁnes a class called a C 0 function.i qi dΩ (2. the ﬁnite element method uses isoparametric elements to construct C 0 functions in Ωh . The case where only the function is continuous.. See [26. Chapter 7]).31) + e=1 Γeq W qn dΓ = 0 ¯ In order for the above integrals to be well deﬁned.30) + e=1 Γeq W qn dΓ = 0 ¯ Introducing the Fourier law the above integral becomes Nel Nel G ≈ Gh = e=1 Nel Ωe W ˙ ρcT − Q dΩ + e=1 Ωe W.CHAPTER 2. Standard element interpolation functions which maintain C 0 continuity are discussed in any standard book on the ﬁnite element method (e.g.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, NI is a speciﬁed spatial function called a shape function for node I, ξ are natural coordinates for the element, xI are values of the coordinates at i node I, T I (t) are time dependent nodal values of temperature, and nel is the number of nodes connected to an element. Standard shape functions, for which all the nodal parameters have the value of approximations to the variable, satisfy the condition

Nel

NI (ξ) = 1

I=1

(2.34)

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

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

CHAPTER 2. INTRODUCTION TO STRONG AND WEAK FORMS

11

Jji =

∂xj ∂ξi

(2.40)

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

=

(2.42)

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

=

(2.43)

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

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

Ωe

where MIJ =

Ωe

NI ρ c NJ dΩ

(2.46)

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

Ωe

(2.47)

where KIJ =

Ωe

NI,i k NJ,i dΩ

(2.48)

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

Ωe Γeq

W qn dΓ = W I FI ¯

(2.49)

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

Ωe

12

NI Q dΩ −

Γe q

NI qn dΓ ¯

(2.50)

**The approximate weak form may now be written as
**

Nel

Gh =

e=1

˙ W I (MIJ T J + KIJ T J − FI ) = 0

(2.51)

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

Nel

˙ (MIJ T J + KIJ T J − FI ) = 0

e=1

(2.52)

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

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

2.6

Implementation of elements into FEAP

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

Ωe

NI,i k NJ,i dΩ

(2.55)

and

. where for example L KIJ = l=1 NI. M.59) Each array is computed for a single element as described in the section of the FEAP Programmer Manual on adding an element. 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.i (ξ l ) k NJ.60) (2. The routine uses basic features included in the FEAP system to generate shape functions. Both steady state and transient solutions are permitted.56) For a transient problem is is necessary to also compute MIJ = Ωe NI ρ c NJ dΩ (2. The listing included in Appendix A summarizes an element for the linear heat transfer problem. perform numerical quadrature. INTRODUCTION TO STRONG AND WEAK FORMS 13 FI = Ωe NI Q dΩ − Γeq NI qn dΓ ¯ (2. etc. is included separately to permit solution of the general linear eigenproblem KΦ = MΦΛ (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 (2.61) which can be used to assess the values of basic time parameters in a problem. hence it needs to compute a residual for the equations (see FEAP User and Programmer Manual for details). The heat capacity array. The sides of the block are assumed to also be fully insulated. FEAP is a general non-linear ﬁnite element solution system. One quadrant of the region is modeled as shown by the mesh in Figure 2.i (ξ l )j(ξ l )wl (2.1.57) The above integrals are normally computed using numerical quadrature.58) where j(ξ) is the determinant of J evaluated at the quadrature point ξ l and wl are quadrature weights.

.2.CHAPTER 2.5 are incorporated into FEAP as a user element and the steady state solution computed.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. INTRODUCTION TO STRONG AND WEAK FORMS 14 Figure 2. The contour of temperatures is shown in Figure 2.

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

2) + Γq W qn dΓ ¯ The derivative of the functional with respect to η now may be constructed using conventional methods of calculus. dG Gη − G0 = lim η→0 dη η 16 (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].3) .i dΩ − Ω W Q dΩ (3. The notion of a variation is associated with the concept of a derivative of a functional (i.i k T. Thus.e. 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. T η ) = Ω η W. a function of functions).1 Derivatives of functionals: The variation The weak form of a diﬀerential equation is also called a variational equation. it is necessary to introduce a scalar parameter which may be used as the limiting parameter in the derivative [10]. the result Gη = G(W. Introducing the family of functions T η into the functional we obtain..Chapter 3 Introduction to Variational Theorems 3. In order to construct a derivative of a functional.

4) dη dη With this result in hand. T ) as dΠ(T η ) = G(τ.7) Symmetry of the inner product resulting from the variation of a weak form is a suﬃcient condition for the existence of a variational theorem which may also be used to generate a weak form.i + ητ. hence the derivative is a variation of G. To investigate symmetry of a functional we consider only terms which include both the dependent variable and the arbitrary function. Thus.CHAPTER 3. often we may only have a . given a functional Π(T ) we can construct G(W. T ) η→0 dη lim (3. 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 . 3. τ ) = A(τ. Thus. For the linear steady state heat equation the derivative with respect to η is constant. given by a functional Π(T ).5) The limit of the derivative as η goes to zero is called the variation of the functional.i k τ. We shall deﬁne the derivative of the functional representing the weak form of a diﬀerential equation as dG = A(W. however. the derivative of the functional with respect to η is given by dG = dη W. The construction of the derivative of the functional requires the computation of variations of derivatives of T .i d = (T. has a ﬁrst variation which is identical to the weak form.8) Note that use of Eq. INTRODUCTION TO VARIATIONAL THEOREMS 17 where G0 is the value of Gη for η equal to 0.6) 3. An inner product is symmetric if A(W.2 Symmetry of inner products Symmetry of inner product relations is fundamental to the derivation of variational theorems. τ ) dη This is a notation commonly used to deﬁne inner products.i dΩ Ω (3. Variational theorems are quite common for several problem classes.i (3. A variational theorem. Using the above deﬁnition we obtain d(T η ). W ) (3. (3.i ) = τ. Symmetry of the functional A also implies that the tangent matrix (computed from the second variation of the theorem or the ﬁrst variation of the weak form) of a Bubnov-Galerkin ﬁnite element method will be symmetric.1 leads to a result where τ replaces W in the weak form.

).i kT.i dΩ − Ω Ω T QdΩ + Γq T qn dΓ ¯ (3.9) (3.CHAPTER 3. A variational theorem is a functional whose ﬁrst variation. t) → ηT (x. T ) W (x) → T (x.g. ηT )dη (3.i dΩ − Ω T Q dΩ + Γq T qn dΓ ¯ (3. stop: no varitational theorem exists. minimum or stationary value.10) Π(T ) = 0 G(T. 2. otherwise. yields the governing diﬀerential equations and boundary conditions associated with some problem.11) Performing the variation of Π and setting to zero gives dΠ(T η ) = G(τ. ηT ) 3.i k η T. Integrate the functional result from (b) with respect to η over the interval 0 to 1. τ ).12) and a problem commonly referred to as a variational theorem. The construction of a variational theorem from a weak form is performed as follows [24]: 1.8. t) T (x. For the steady state heat equation we have G(T. Check symmetry of the functional A(W. existence of a variational theorem yields a weak form directly by using Eq. In practice we seldom need to have the variational theorem.13) The integral is trivial and gives Π(T ) = 1 2 T. Perform the following substitutions in G(W. but knowledge that it exists is helpful since it implies properties of the discrete problem which are beneﬁcial (e..14) . when set to zero. etc. The result of the above process gives 1 (3. t) to deﬁne G(T. Also. ηT ) = Ω T. INTRODUCTION TO VARIATIONAL THEOREMS 18 functional G and desire to know if a variational theorem exists. T ) = 0 η→0 dη lim (3. 3. symmetry of the tangent matrices. If symmetric then to to 2.

19) Letting tn+1 − tn = ∆t and omitting the subscripts for quantities evaluated at tn+1 .20) (3. we ﬁrst discretize the transient term using some time integration method. A is positive for all τ ).i dΩ (3. T ) ˙ (3.18) If however. The second variation of the theorem generates the inner product A(τ. τ ) (3.e. 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. 3. the rate term which includes both T and τ becomes T .T ∆t A= = η (3.21) .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 . the variational theorem is called a minimum principle and the discrete tangent matrix is positive deﬁnite.16) If the second variation is strictly positive (i. ητ ∆t τ .i k T.CHAPTER 3. INTRODUCTION TO VARIATIONAL THEOREMS 19 Reversing the process.15) and performing the derivative deﬁned by Eq.12. we can often restore symmetry to the functional and then deduce a variational theorem for the discrete problem. the ﬁrst variation of the variational theorem generates a variational equation which is the weak form of the partial diﬀerential equation. ητ = (η τ . The transient heat equation with weak form given by G = Ω W ˙ ρ c T − Q dΩ + Ω W.. 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 ﬁrst variation is deﬁned by replacing T by T η = T + ητ (3.

25) Performing the derivatives leads to δΠ = 1 2 (k T. 3.3 Variational notation A formalism for constructing a variation of a functional may be identiﬁed and is similar to constructing the diﬀerential of a function.26) Collecting terms we have δΠ = Ω δT.i ∂ (T qn ) δT dΓ ¯ Γq ∂T Ω ∂ (T Q) δT dΩ ∂T (3. For the functional Eq.i are the dependent variables of the functional. That is.3).i k T. 3.i dΩ − Ω Ω Q δT dΩ + Γq qn δT dΓ ¯ (3. The diﬀerential of a function f (xi ) may be written as ∂f df = dxi (3.e.i dΩ − Ω ∂T. it is formally the ητ (x)). etc.i dΩ − Ω Q δT dΩ + Γq qn δT dΓ ¯ (3.i ) δT. τ ) (3.i k T. We shall address this aspect at a later time.i + · · · (3. INTRODUCTION TO VARIATIONAL THEOREMS 20 since scalars can be moved from either term without aﬀecting the value of the term. A = (T. u. This formal construction is easy to apply but masks the meaning of a variation.22) 3.i where u.2 with δT replacing W . δu is the variation of the variable (i.i k) δT. However.24) ∂u ∂u.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. η τ ) = (η T.CHAPTER 3. Similarly.27) which is identical to Eq. and δΠ is called the ﬁrst variation of the functional..23) ∂xi where xi are the set of independent variables. . we may formally write a ﬁrst variation as ∂Π ∂Π δΠ = δu + δu. applying this construction can be formally performed using usual constructions for a derivative of a function. 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.i + T. we obtain the result δΠ = + 1 2 ∂ (T.

For a presentation using indicial notation see [26. σ.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. u. (4. (4. and u is the acceleration. . x.2) u. Deformation measures based upon the gradient of the displacement vector. and the strain tensor.4) . Balance of linear momentum expressed as ¨ · σ + ρ bm = ρ u where ρ is the mass density. which leads to symmetry of the stress tensor σ = σT 3. The basic governing equations are: 1.3) u + (a) u u = 1 2 21 u + ( u)T (4. where the domain of analysis is Ω with boundary Γ. The presentation below assumes small (inﬁnitesimal) deformations and general three dimensional behavior in a Cartesian coordinate system. Chapter 6]. The dependent variables are given in terms of the displacement vector. Balance of angular momentum. the stress tensor. ¨ operator. which may be split as follows u = where the symmetric part is (s) (s) (4. bm is the body force per unit mass.1) is the gradient 2.

3 u3.2 u2. ω13 0 ω12 = −ω12 ω23 0 ω33 −ω13 −ω23 (4.8) σ31 σ32 σ33 which from the balance of angular momentum must be symmetric.3 u2. hence σij = σji The gradient of the displacement has the metries) u1.3 (4. the symmetric part deﬁnes the strain = (s) u (4.5) Based upon this split. if the tensor is symmetric only 6 are independent and if the tensor is skew symmetric only 3 are independent.11) and the symmetry condition ij = ji (4.CHAPTER 4.14) .thus. ωij = ωji which implies ω11 = ω22 = ω33 ω11 ω12 ω21 ω22 ω → ω31 ω32 = 0.1 u → u3. SMALL DEFORMATION: LINEAR ELASTICITY and the skew symmetric part is (a) 22 u = 1 2 u − ( u)T (4. The component ordering for each of the tensors is given by σ11 σ12 σ13 σ → σ21 σ22 σ23 (4. However.12) The spin tensor is skew symmetric.13) ω13 ω23 0 (4. or small rotation.2 u3.9) components ordered as (with no sym u1.7) In a three dimensional setting the above tensors have 9 components. Accordingly.1 u2.10) The strain tensor is the symmetric part with components 11 12 22 32 13 23 → 21 31 33 (4. ω = (a) u (4.1 (4.2 u1.6) and the skew symmetric part deﬁnes the spin.

Γ. Inclusion of intertial forces requires the speciﬁcation of the initial conditions u(x. We shall consider several alternatives for splits during the course. t In the balance of momentum.18) ˙ u(x.21) . For this purpose the boundary. and v0 is the initial velocity ﬁeld.15) Static or quasi-static problems are considered by omitting the acceleration term from the momentum equation (Eq. For example. 0) = v0 (x) where d0 is the initial displacement ﬁeld. and • speciﬁed tractions on the part Γt .CHAPTER 4. given as: ¯ u = u ¯ where u is a speciﬁed quantity. 0 as a thermal strain is given by 0 = th = α(T − T0 ) (4. (4. In small deformation analysis the strain is expressed as an additive sum of parts. 0) = d0 (x) (4.19) 4. Accordingly. SMALL DEFORMATION: LINEAR ELASTICITY 23 The basic equations which are independent of material constitution are completed by specifying the boundary conditions. 0 is a second part which we assume is a speciﬁed strain.. given as: t = σn = ¯ t where ¯ is a speciﬁed quantity. This may be converted to a body force per unit volume (i. = m + 0 (4.1). unit weight/volume) using ρ bm = bv (4. the body force was speciﬁed per unit of mass.e. is split into two parts: • Speciﬁed displacements on the part Γu .16) (4.1 Constitutive Equations for Linear Elasticity The linear theory is completed by specifying the constitutive behavior for the material. 4.17) (4.20) where m is the strain caused by stresses and is called the mechanical part. however. we begin by considering a linear elastic material with an additional known strain.

C1233 → D43 . we have C1111 → D11 .1 Thus. the constitutive equation for linear elasticity is written in indicial notation as: σab = Cabcd ( cd − 0 ) (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 .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.CHAPTER 4.22) where the matrix of stresses is ordered as the vector σ = σ11 σ22 σ33 σ12 σ23 σ31 T (4. using this table.23) the matrix of strains is ordered as the vector (note factors of 2 are used to make shearing components the engineering strains.28) cd The transformation from the tensor to the matrix (Voigt) form is accomplished by the index transformations shown in Table 4. etc. (4.LP where T is temperature and T0 is a stress free temperature. W ( computed as ∂W σab = ∂ m ab the elastic modulus matrix is symmetric and satisﬁes Dij = Dji ). 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. SMALL DEFORMATION: LINEAR ELASTICITY 24 .27) Using tensor quantities. from which stresses are (4. γij ) = 11 22 33 2 12 2 23 2 T 31 (4.25) Assuming the existence of a strain energy density.26) (4.

Later. we can add the inertial eﬀects and use time discrete methods to restore symmetry to the formulation.CHAPTER 4. we assume that inertial forces may be ignored. . For the present. 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 4.1: 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.

a transpose may be introduced into each term without changing the meaning.2) A variational theorem is stationary when the arguments (e. K. and. E.. as done for the thermal problem. thus.g. S. I = Ω aT b dΩ = Ω (aT b)T dΩ = Ω bT a dΩ (5. ) = + Ω 1 2 T Ω D dΩ − Ω (s) T D 0 dΩ σT ( u − ) dΩ − Ω uT bv d Ω (5.3) . The theorem may be written as I(u. u. Accordingly. we proceed as in the previous chapter. is known as the Hu-Washizu variational theorem. we introduce the variations to the displacement. ) satisfy the conditions where the ﬁrst variation vanishes. hence. To construct the ﬁrst variation. as uη = u + η U 26 (5. Washizu [25]. The variational theorem is a result of the work of the Chinese scholar. and the strain. U. the stress.1 Hu-Washizu Variational Theorem Instead of constructing the weak form of the equations and then deducing the existence of a variational theorem.1) − Γt uT ¯ dΓ − t Γu ¯ tT (u − u)dΓ = Stationary Note that the integral deﬁning the variational theorem is a scalar. σ. a variational theorem which includes all the equations for the linear theory of elasticity (without inertial forces) will be stated.Chapter 5 Variational Theorems: Linear Elasticity 5. For example. σ. Hu. and the Japanese scholar.

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

we can assume that it is equal to the value of the non-vanishing expression. the balance of linear momentum. This leads to a contradiction. the variations multiply the constitutive equation. On the other . consequently.11 is not possible in general.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. the strain-displacement equation. Suppose that an expression does not vanish at a point. Two of these. 5. u.CHAPTER 5. For linear elasticity this leads to = 0 + D−1 σ (5.2 Hellinger-Reissner Variational Theorem The Hellinger-Reissner principle eliminates the strain as a primary dependent variable. For the Hu-Washizu theorem. and the stress. then. 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. σ. the traction boundary condition. only the displacement.11) The need to develop an expression for strains in terms of stresses limits the application of the Hellinger-Reissner principle. in ﬁnite deformation elasticity the development of a relation similar to 5. which must then be positive. the only equation not contained is the balance of angular momentum. This results in the integral of the square of a function. The Hu-Washizu variational principle will serve as the basis for most of what we need in the course. and hence the integral will not be zero. There are other variational principles which can be deduced directly from the principle. and thus the only possibility is that the assumption of a non-vanishing expression is false. The lemma is easy to prove. For example. since the variation is arbitrary. The strains are eliminated by developing an expression in terms of the stresses. and the displacement boundary condition. 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. remain as arguments in the functional for which variations are constructed. The expression which multiplies each variation function is called an Euler equation of the variational theorem. Indeed.

The ﬁrst term in 5. σ. The MPE theorem may be deduced by assuming = (s) u (5.11 into the Hu-Washizu principle leads to the result I(u.15 and 4. 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. Introducing 5.g. 4.CHAPTER 5.15) are satisﬁed at each point of Ω and Γ. respectively.14 and 5.14) and ¯ u = u (5.12 may be omitted since its ﬁrst variation is zero. and the strain.11 to 5.15 the relation σ = D( − 0 ) (5.13) together with 4.16) uT bv dΩ − Γt uT ¯ tdΓ Since stress does not appear explicitly in the theorem. we shall on occasion use the principle in our studies. In addition.1. The strain-displacement equations are deduced by either directly stating 4. . the Hellinger-Reissner principle is an important limiting case when considering problems with constraints (e. 5. thin plates as a limit case of the thick Mindlin-Reissner theory). Accordingly.12) − Γt u tdΓ − Γu T¯ ¯ t (u − u)dΓ The Euler equations for this principle are (s) u = 0 + D−1 σ (5. Thus. Thus. the constitutive equation must be given.13. as arguments of the functional.3 Minimum Potential Energy Theorem The principle of minimum potential energy eliminates both the stress.16. the displacement boundary conditions are assumed to be imposed as a constraint on the principle.17) . 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. linear elastic incompressible problems. VARIATIONAL THEOREMS: LINEAR ELASTICITY 29 hand. in addition to 5.6 or comparing 5..

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

As indicated in Chapter 2 the region is divided as Mel Ω ≈ Ωh = e=1 Ωe (6.1) and integrals are deﬁned as Mel ( · ) dΩ ≈ Ω Ωh ( · ) dΩ = e=1 Ωe ( · ) dΩ (6. ξ are natural coordinates for the element. This reduces a steady state problem to an algebraic process and a transient problem to a set of ordinary diﬀerential equations. uα are the values of the displacement vector at node α and repeated indices imply summation over the range of the index.2) In the above Mel is the total number of elements in the ﬁnite element mesh. Using the isoparametric concept 31 . The ﬁnite element approximation for displacements in an element is introduced as Nel u(ξ. In the ﬁnite element method we divide the region into elements and perform the approximations on each element. A similar construction is performed for the boundaries.3) where Nα is the shape function at node α. t) = α=1 Nα (ξ) uα (t) = Nα (ξ) uα (t) (6. In the direct method of the calculus of variations the dependent variables are expressed as a set of trial functions multiplying parameters.Chapter 6 Displacement Finite Element Methods A variational equation or theorem may be solved using the direct method of the calculus of variations. With this construction the parts of the variational equation or theorem are evaluated element by element.

DISPLACEMENT METHODS 32 x(ξ) = Nα (ξ) xα (6. 6. and the applied surface tractions.7 yields Πf = (uα )T Ωe Nα bv dΩ + Γte Nα ¯ dΓ t = (uα )T Fα (6.9) 6.4) where xα are the cartesian coordinates of nodes.5) and set ¯ ¯ uα (t) = u(xα . Πf = Ωe uT bv dΩ + Γte uT ¯ dΓ t (6. bv . the displacement at each point in an element may be computed. we express ¯ ¯ u = Nα (ξ) uα (t) (6.7) Using Eq.CHAPTER 6. ¯ The terms in the variational principal are t.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. of the term Πσ = Ωe ( (s) u)T σ dΩ = e Ωe ( (s) u)T σ dΩ (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. The remaining terms involving speciﬁed applied loads are due to the body forces. u.10) . 6. t) (6. In the next sections we consider the computation of the external force (from applied loads) and the internal force (from stresses) by the ﬁnite element process. This step is not necessary but is common in most applications. 6.3 in Eq.6) We then will assume the integral over Γu is satisﬁed and may be omitted.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. That is.

1 u2.15) = 11 22 33 2 12 (6.11) where Bα is the strain displacement matrix for the element. DISPLACEMENT METHODS 33 Using the ﬁnite element approximation for displacement.1 + u1.1 ) (u2.16) and are expressed in terms of the displacement derivatives as T = u1.2 0 0 Nα.1 u2.1 ) (6. for a 2-dimensional axisymmetric problem (with no torsional loading) the strains are T = (6.1 0 0 Bα = Nα.2 0 Nα.2 = 0 0 Nα.1 u2.2 u1 /x1 (u1. 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 (6.18) Finally.3 ) as: 0 0 Nα.2 0 Nα.2 + u2.1 0 Nα.1 0 0 Nα. If the components of the strain for 3-dimensional problems are ordered as T = 11 22 33 2 12 2 23 2 31 (6. Bα becomes: Bα (6.12) and related to the displacement derivatives by T = u1.14) ∂Nα ∂xi For a 2-dimensional plane strain problem the non-zero strains reduce to Nα.3 Nα.2 Nα.19) 11 22 33 2 12 and are expressed in terms of the displacements as T = u1.3 + u3.3 (u1.2 ) (u3.3 where (6.2 + u2.2 + u2.1 ) Nα.2 u3.1 (6.13) the strain-displacement matrix is expressed Nα.CHAPTER 6.1 (6.17) thus.20) .i = T (6.3 Nα.2 0 (u1.

1 0 0 Nα. 6. The stress may be written in terms of the deviatoric and pressure parts (pressure is spherical part) as σ = s + p1 (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. For stress the spherical part is the mean stress deﬁned by 1 1 (6.x2 plane. Locking is generally associated with poor performance at or near the incompressible limit. the internal stress divergence force. becomes: Nα.26) The deviatoric part of stress . DISPLACEMENT METHODS The strain-displacement matrix for axisymmetry. is deﬁned so that its trace is zero. Bα . .23) which gives Πσe = (uα )T Pα (σ) (6.21) (Bα )T σ dΩ (6.22) In the sequel we deﬁne the variation of this term with respect to the nodal displacements. s.27) 1 For axisymmetry it is also necessary to replace the volume element by dΩ → x1 dx1 dx2 and the surface element by dΓ → x1 dS where dS is an boundary diﬀerential in the x1 . uα . To study the locking problem we split the formulation into deviatoric and volumetric terms. z.2 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. This force is expressed by Pα (σ) = Ωe (Bα )T σ dΩ (6.2 Bα = Nα /x1 0 Nα. respectively1 The stress divergence term for each element may be written as Πσe = (uα )T Ωe 34 (6.1 where x1 .CHAPTER 6. x2 now denote the axisymmetric coordinates r.25) p = tr(σ) = σkk 3 3 For inﬁnitesimal strains the spherical part is the volume change deﬁned by θ = tr( ) = kk (6.

Utilizing the above properties. which in matrix notation is given by the vector mT = 1 1 1 0 0 0 (6. For subsequent developments.38) In the above m is any positive integer power.31) and hence mT s = 0 as required.29) thus. that inverses to the projectors do not exist. however.39) (6.35) (6. DISPLACEMENT METHODS 35 where. the volumetric projector is deﬁned by Ivol = 1 m mT 3 (6.30) where.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. 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 deviatoric and volumetric parts are given by = e + 1 θ1 3 (6. in three dimensions. we can operate on the strain to deﬁne its deviatoric and volumetric parts.32) as the deviatoric projector. Similarly.33) (6.37) . we deﬁne Idev = I − 1 m mT 3 (6. 1 is the rank two identity tensor.36) (6. Accordingly. I is a 6 × 6 identity matrix. We note that the trace of the stress gives mT σ = 3 p = mT s + p mT m = mT s + 3 p (6. We note.28) In matrix form the pressure is given by p = 1 T m σ 3 (6.CHAPTER 6.

1 3 Nα.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 −Nα.2 −Nα.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 + Nα /x1 Nα.3 −Nα.1 2 Nα. Accordingly.44) (6.1 Nα.3 (6.50) . DISPLACEMENT METHODS 36 where e is the strain deviator and θ is the change in volume.2 3 Nα.3 (6.2 (6. In 2dimensional plane problems the volumetric strain-displacement matrix is given by bα = Nα.1 Nα.2 3 Nα. mT e = 0 (6.2 Nα.2 3 Nα.1 (6.1 0 3 Nα.1 = 3 −Nα.3 3 Nα.47) and for 2-dimensional axisymmetric problems bα = Nα.46) mT Bdev = 0 (6.2 −Nα.40) we obtain e = Idev .1 −Nα. 6. Using matrix notation we have θ = mT (6.1 −Nα.45) 1 mb 3 (6.43) (6.3 1 −Nα.2 (6.CHAPTER 6.48) The deviatoric matrix Bdev is constructed from Eq.1 1 −Nα. For 3-dimensional problems bα = Nα.39 and yields for the 3-dimensional problem 2 Nα.42) is the volumetric strain-displacement matrix for a node α in its basic form.2 2 Nα.49) Bdev = 0 3 3 Nα.3 0 3 Nα.

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 (6.2 (6. the internal force is composed of the sum of deviatoric and volumetric parts. 6.1 − Nα /x1 ) −Nα.CHAPTER 6.1 ) −Nα.2 3 Nα. DISPLACEMENT METHODS 37 Finally.4 Internal Force .57) .56) Thus.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. Pα = Ωe BT σ dΩ = α Ωe BT (s + p m) dΩ α (6.52) which after rearrangement gives Pα = Ωe BT s dΩ + α Ωe BT m p dΩ α (6.2 1 −(Nα.51) Bdev = 3 (2 Nα /x1 + Nα. then B = Bdev + Bvol = Bdev + Pα = Ωe (BT )α s dΩ + dev Ωe bT p dΩ α (6. Accordingly.53) If we introduce 1 mb (6. the deviatoric matrix for the 2-dimensional axisymmetric problem is given by: (2 Nα.2 3 Nα.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 ) 2 Nα.1 6.

the Lam´ parameters have the following ranges e 0 ≤ ν ≤ and 1 2 . K.61) 1 We note that K also tends to inﬁnity as ν approaches 2 .67) .1 and expressed as σ = D (6. and λ is a parameter which causes diﬃculties since it is inﬁnite. and e Poisson’s ratio.59) 1 E E . Another parameter which is related to λ and µ is the bulk modulus.62) (6.CHAPTER 6.60) 2 2 3 1 For an incompressible material ν is 2 . which is deﬁned by 0 ≤ ν ≤ K = λ + E 2 µ = 3 3 (1 − 2 ν) (6. For an isotropic material the elastic moduli are then related by cijkl = λ δij δkl + µ (δik δjl + δil δjk ) (6. µ = E 2 (1 + ν) (6. ν.58) For diﬀerent values of ν. 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 (6.64) We note that the above deﬁnition for the moduli satisﬁes all the necessary symmetry conditions. that is cijkl = cklij = cjikl = cijlk (6. DISPLACEMENT METHODS 38 where λ and µ are the Lam´ parameters which are related to Young’s modulus. by λ = νE (1 + ν)(1 − 2 ν) . 0 ≤ λ ≤ ∞ (6.63) where cijkl are the elastic moduli.65) The relations may be transformed to matrix (Voigt) notation following Table 4. E. ≥ µ ≥ (6.66) where the elastic moduli are split into D = λ Dλ + µ Dµ (6.

thus. the computation of the element stiﬀness matrix for a displacement approach is given and is based upon the above representations for the moduli.e.72) Dµ Idev = Idev Dµ (6.73) Once Ddev has been computed it may be noted that Idev Ddev = Ddev Idev = Ddev Ivol Ddev = Ddev Ivol = 0 and.70) (6.68) Dµ (6.2 If the moduli matrices are premultiplied by Ivol and Idev the following results are obtained Ivol Dλ = Dλ Idev Dλ = 0 2 2 m m T = Dλ Ivol Dµ = 3 3 and 4 −2 −2 0 0 0 −2 4 −2 0 0 0 1 −2 −2 4 0 0 0 = Ddev = 0 0 3 0 0 3 0 0 0 0 0 3 0 0 0 0 0 0 3 (6.75) .71) (6. γij = 2 ij ).74) (6.. Note that in Dµ the terms multiplying shears have unit values since engineering shear strains are used (i. it is a deviatoric quantity.CHAPTER 6.69) used as non-dimensional matrices to split the moduli. 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. 2 (6. In the following section.

6 Stiﬀness for Displacement Formulation The displacement formulation is accomplished for a linear elastic material by noting that the constitutive equation is given by (for simplicity 0 is assumed to be zero) σ = D The strains for a displacement approach are given by = Bβ uβ where uβ are the displacements at node β. DISPLACEMENT METHODS 40 6.85) 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 (6.78) (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 ) (6.76) 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.CHAPTER 6.83) (6.81) = Idev (λ Dλ + µ Dµ ) (6.2 that Pα = Ωe (6.80) 2 µ) Dλ = K Dλ 3 Thus.79) If we use the properties of the moduli multiplied by the projectors.86) . the above equations reduce to s = µ Ddev = µDµ e = µDµ (Bdev )β uβ (6.84) (BT )α s dΩ + dev Ωe bT p dΩ α (6.82) (6.77) (6. the pressure constitutive equation is p m = (λ + and = K m(mT ) = K m θ (6.

87) = Ωe µ (BT )α Dµ (Bdev )β dΩ = dev Ωe µ BT Ddev Bβ dΩ α (6. The next lowest order of quadrature is called a reduced quadrature.. 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. 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. use of standard quadrature on one term and reduced quadrature on another leads to a method called selective reduced quadrature.92) f (x(ξ)) j(ξ) dξ ≈ l=1 f (x(ξ l )) j(ξ l ) wl (6.e. quadrature).93) . DISPLACEMENT METHODS 41 and.CHAPTER 6.91) (6. thus. 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. 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. for isotropic linear elasticity.89) 6. Alternatively.90) where 2 denotes integration over the natural coordinates ξ. dξ denotes dξ1 dξ2 in 2-dimensions.88) K bα bT dΩ = β Ωe K BT Dλ Bβ dΩ α (6. 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.

98) . a shape function routine for 2 dimensions is called shp2d and is accessed by the call (6. for 2-dimensions. Loop over the quadrature points. For example. For example. • Accumulate the array. l • Compute j(ξ l ) wl = c • Compute the matrix in the integrand. The evaluation of the shape functions is performed using a shape function subprogram.. DISPLACEMENT METHODS 42 where ξ l and wl are quadrature points and quadrature weights. (e. Additional steps are involved in computing the entries in each array. In FEAP. Pα ← Pα + Aα c 3.96) 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. Repeat step 2 until all quadrature points in element are considered.g. The use of quadrature reduces the development of ﬁnite element arrays to an algebraic process involving matrix operations.97) Similar expressions may be deduced for each of the terms deﬁned by the deviatoric/volumetric splits. the determination of Bα requires computation of the derivatives of the shape functions. the integration is generally carried out as a product of one-dimensional Gaussian quadrature. respectively.95) Using quadrature. the stress divergence is given by L Pα = l=1 Bα (ξ l )T σ(ξ l ) j(ξ l ) wl (6.g. the basic algorithm to compute the stress divergence term is given by: 1. Initialize the array Pα 2.. e. Nα.i . and computation of σ l requires an evaluation of the constitutive equation at the quadrature point. Bα (ξ l )T σ l = Aα ). Thus. 1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 (6.CHAPTER 6.94) and for 3-dimensions 1 1 −1 1 g(ξ) dξ = 2 −1 −1 g(ξ) dξ1 dξ2 dξ3 (6. For brick elements in three dimensions and quadrilateral elements in two dimensions.

1 shp(2.2 shp(3. l The array of points and weights has the following meanings: .A) is NA. shp. lint -total number of quadrature points (output). if false derivatives returned with respect to x (input). ndm.3 The quadrature points may be obtained by a call to int2d: call int2d( l. swg ) where -number of quadrature points in each direction (input). ξ2 ) at quadrature point (input) xl array of nodal coordinates for element (xl(ndm. if true derivatives returned with respect to ξ. flag) where xi natural coordinate values (ξ1 . swg -array of natural coordinates and weights (output).CHAPTER 6.nen)) (input) shp array of shape functions and derivatives (shp(3. 43 The array of shape functions has the following meanings: shp(1. DISPLACEMENT METHODS call shp2d( xi. lint. xsj. nel. xl.A) is NA. ix.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.A) is NA.

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

4) (t) 45 . σ.Chapter 7 Mixed Finite Element Methods 7.2) (7. displacements appear up to ﬁrst derivatives. In the work considered here we use the Hu-Washizu variational principle. while the stresses and strains appear without any derivatives. Accordingly.1) In the principle. the continuity conditions we may use in ﬁnite element approximations are C 0 for the displacements and C −1 for the stresses and strains (a C −1 function is one whose ﬁrst integral will be continuous). Appropriate interpolations for each element are thus u(ξ) = NI (ξ) uI (t) σ(ξ) = φα (ξ) σ α (t) and (ξ) = ψα (ξ) α (7. which we recall may be written as Π(u. ) = + Ω 1 2 T Ω D dΩ − Ω (s) T D 0 dΩ uT bv dΩ σT ( T u − ) dΩ − Ω T − Γt u ¯ dΓ − t Γu ¯ t (u − u) dΓ = Stationary (7.1 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.3) (7.

12) (7. If. respectively.11) (7. φα and ψα need only be piecewise continuous in each element.13) Kθ − p = 0 (7. σ. however.8) 1 Strictly. 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. ) = + Ω 1 2 µ Ω T Ddev dΩ − Ω µ T Ddev 0 dΩ (7. for the present.10) (7. this makes the evaluation of integrals over each element more diﬃcult and to date is rarely used. Strain-Displacement equations e(u) − e = 0 θ(u) − θ = 0 3. and the traction and displacement boundary integrals and consider an isotropic linear elastic material.6) (7. we ignore the integral for the body force.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. Constructing the variation for the above split leads to the following Euler equations which hold in the domain Ω: 1. the remaining terms may be split into deviatoric and volumetric parts as Π(u.CHAPTER 7.7) u) = are the strain-displacement relations for the deviatoric and volumetric parts. may have no direct physical meaning.9) (7. . Constitutive equations µ Ddev − s = 0 (7.1 The parameters σ α and α are not necessarily nodal values and. thus. Balance of Momentum · (s + 1 p) + bv = 0 which is also written as div(s + 1 p) + bv = 0 2.

14) If the deviatoric part is approximated by taking e = e(u) (7.15) for each point of Ω. 7.13 at each point of the domain of an element. If the material is isotropic linear elastic. ) ≈ Πh (u. we assume p(ξ) = φα (ξ) pα (t) = φ(ξ) p θ(ξ) = φα (ξ) θα (t) = φ(ξ) θ (7. The variational expression Eq. 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 (7.19) where it is noted that the same approximating functions are used for both p and theta. p. p.16) p [θ(u) − θ] dΩ which may be split into integrals over the elements as Π(u. this part of the problem is given as a displacement model.21) . ) = e Πe (u.18) (7. σ. MIXED FINITE ELEMENT METHODS In addition the boundary conditions for Γu and Γt are obtained. σ.5 becomes Π(u. For other situations.17) A mixed approximation may now be used to describe the pressure and the volume change in each element.CHAPTER 7. p. the constitutive equation may be approximately satisﬁed. σ. p.20) Thus. θ) (7. θ) = e Πe (u. the use of the same functions will permit an exact satisfaction of the constitutive equation. the ﬁnite element approximation for the mixed formulation may be written as Πe (u. θ) ≈ Πh (u. the Hu-Washizu variational theorem may be approximated by summing the integrals over each element. Accordingly. p. Eq. 47 Using the interpolations described above. Recall that the strain-displacement equations for a ﬁnite element approximation are given by (u) = BI uI (7. Accordingly. 7. Π(u. 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. ) (7.

31) . The mixed variational terms become Πe (u.24) (7. 7.27) where e0 are the deviatoric initial strains.CHAPTER 7. Θ dη 0 −h k θ 0 (Pdev )I 0 − π0 (7. θ) = (uI )T + θT + pT 1 (Kdev )IJ uJ − (P0 )I dev 2 1 k θ − π0 2 g J uJ − h θ (7. p.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 (7.28 may be written in the matrix form J (Kdev )IJ gJ 0 u dΠe T T I T T p gI 0 −h = (U ) .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.30) (7.29) (7. Π .23) (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. MIXED FINITE ELEMENT METHODS If we deﬁne the following matrices: k = Ωe 48 K φT φ dΩ K φT θ0 dΩ Ωe (7.22) π0 = h = Ωe (7.

37) ¯ ¯ (Kdev )IJ + bT k bJ I ¯ uJ − (P0 )I − bT π 0 dev I (7.36) (7. from the stationarity condition. since we used the same functions for the two approximations. 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 ) . the matrix h is square and positive deﬁnite (provided our approximating functions are linearly independent). by deﬁning a modiﬁed volumetric strain-displacement matrix as ¯ bI = h−1 gJ the above simpliﬁes to dΠe = (UI )T dη 2 (7. The requirement for a solution to exist is that2 nθ ≥ np (7. . consequently. We have satisﬁed this requirement by taking an equal number for the two approximations. we may perform the element solutions by inverting only h. consequently. respectively. δθ 0 −h k θ 0 (Pdev )I 0 − π0 49 (7. The solution to Eq.35) (7. 7. the last two rows of the above matrix expression must vanish and may be solved at the element level.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. Also.33) where nθ and np are the number of parameters associated with the volume change and pressure approximations. Chapter 12].34) Finally. See [26.38) This is a mixed patch test requirement.32) We note that the parameters p and θ (and their variations Π and Θ) are associated with a single element.CHAPTER 7. δp .

41) where the deviatoric stress is expressed by the displacement approximation as s = µ Ddev (BJ uJ − 0 ) (7. The number of φ functions will aﬀect the rank of the volumetric terms. and the number is small compared to the number of degrees-of-freedom on the element.45) for the approximating space.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.43) 7.40) BT (s + p m) dΩ I (7.44) Provided the approximations for φ are linearly independent. φ2 = ξ1 (7. rank(k) ) (7.39) (7. 4-node quadrilateral or 8-node brick elements can use a single function φ1 = 1 (7. For example. 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. For example.46) . use of φ1 = 1 . if we show that two functions are suﬃcient for a 2-dimensional element. the rank will normally be that of k.42) and the pressure is expressed by the mixed approximation as p = φ(ξ) h−1 (k θ − π 0 ) (7.CHAPTER 7. 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. The modiﬁed volumetric stiﬀness has a rank which is given by ¯ rank(Kvol ) = min ( rank(b).

l) (b) Compute the volume element times the quadrature weight jl wl = dv(l) 3. φ2 = ξ1 . Loop over quadrature points.49) (7. An algorithm to implement the above mixed model for linear elasticity where D is constant in each element may be summarized as: 1. B.54) . FEAP will initialize K and the element residual. 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 . l) NI . I. and strains. 2. Several alternatives are possible. gI and h. i(ξ l ) = shp(i. Loop over quadrature points. another is to use 3 functions with φ1 = 1 .50) (7. φ3 = ξ2 (7. Initialize arrays: gI . one being φ1 = 1 . k. Invert h and compute bI 5.53) (b) Compute quadrature stresses and π 0 σl = D ( l − 0 ) (7. NI (ξ l ) = shp(3. l (a) Compute shape functions: In 2-d problems FEAP uses. π 0 .48) φ2 = ξ1 + ξ2 (7. I. Loop over quadrature points. l (a) Compute the volumetric strain matrices. l (7.51) ¯ bI = h−1 gI (7.52) = [Idev BI (ξ l ) + 1 ¯ m φ(ξ l ) bI ] uI 3 (7. l (a) Compute strain-displacement matrix. ¯ 4.CHAPTER 7.47) The actual functions selected must be subjected to further evaluations to decide which best meets the objectives of the problem solution. h.

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 (7.56) 7. is computed from a mixed form.61) (7. we have σ = D[ − 0 ] (7. are computed directly from the displacements but the volumetric strain. k 6. ¯ = 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 (7.57) 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. 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. Kdev (e) Compute the volumetric local tangent.CHAPTER 7.3 Mixed Solutions for Anisotropic Linear Elastic Materials A more general form of the Hu-Washizu principle is needed to consider either anisotropic linear elastic materials or inelastic materials in which there is coupling between volumetric and deviatoric eﬀects. The stress may be split into deviatoric and pressure parts as ¯ σ = ¯ + mp s ¯ (7.58) ) + (7.62) ) + . Accordingly.55) (d) Compute the deviatoric tangent.59) . Accordingly. K ¯ ¯ K = Kdev + bT k b (7. Compute the tangent. e. θ. In this section we construct the form of the functional for an anisotropic linear elastic material.

adding the terms associated with the mixed volumetric pressure and volume change Vainberg’s theorem may be used to obtain a variational theorem.19. the ﬁrst variation of 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 (7.63) (7. the stress and strain splits may be substituted into 7. p.1. For isotropy d is zero.66) (7.18 and 7. Alternatively. 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 (7.65) (7.70) ˆ dΩ + δ pT T Ωe −φ φ bT φ I T ˆ dΩ p + δIext .69 are all associated with d which deﬁnes a coupling between deviatoric and volumetric strains.64) (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.CHAPTER 7. The added terms in 7.69) This form of the variational principle is equivalent to 7. If we introduce ﬁnite element interpolations using standard displacement interpolation together with the pressure and volume interpolations given by 7. Subsequently.16 which was deduced for isotropic materials.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 result is Π(u.67) − dvol θ0 (7.

73) where h and gI are as deﬁned in 7. however. yields ˆI δΠσ = δ uT Ω BT [¯ + m p] dΩ I s (7. The stress σ..70 may be expressed in terms of stresses by substituting the interpolations into 7. similarly. bI .61 and 7. the equation ˆ multiplying δ θ yields the equation φT p dΩ = ¯ Ωe Ωe ˆ ˆ φT φ dΩ p = h p (7. and the mixed volumetric ¯ strain displacement equation.77) ¯ ¯ and. when we later consider other material models (e. viscoelasticity. MIXED FINITE ELEMENT METHODS 54 The variational equation 7. ˆ the multiple of δ p yields ˆ φT bJ dΩ uJ = Ωe Ωe ˆ ˆ φT φ dΩ θ = h θ (7.62 resulting in δΠe = ˆI δ uT ˆT δθ Ωe BT ¯ Is dΩ φT p ¯ ˆ uJ ˆ θ (7. Thus. etc.) the eﬀective material moduli are the ones computed by linearizing the constitutive equation expressed in terms of the .76) The stress of the mixed method is deﬁned as σ = ¯ + mp s (7. is the stress which is computed from the constitutive equation for each material. Accordingly.CHAPTER 7.72) which yields ˆ ¯ ˆ ˆ θ = bI uI = h−1 gI uI (7. 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. p. is not equal to σ. in general. respectively.74) Using these results.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.g.24 and 7. plasticity.75) which upon use of the deﬁnitions for the mixed pressure.25.

The tangent matrix may be expressed in terms of the displacements alone by writing the variational equation 7. Constitution computation for each quadrature point (a) Compute ˆ = BI uI ¯ ˆ θ = φ(ξ) bI uI ¯ = Idev + 1 mθ 3 0 (7. Mixed volumetric strain displacement matrix ¯ (a) Compute bI = h−1 gI 3. ξ1 .81) (7. · · · ] (for the 4-node element φ = 1 (b) Compute arrays h = Ωe φT φ dΩ φT bJ dΩ Ωe (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.80) The algorithm for the development of a mixed element based upon the above may be summarized as: 1.74 the dependence of 7.78 on θ and p may be eliminated to give ˆI ¯ ˆ δΠe = δ uT KIJ uJ + δΠ0 + δΠext (7.84) (7.86) ¯ σ = D[ ¯ − ] .CHAPTER 7.83) (7.70 as ˆ uI (Kdev )IJ kI gJ ˆ ˆT kvol − h θ δΠe = δ uT δ θ δ pT kT ˆI ˆ J T gI −h 0 ˆ p + δΠ0 + δΠext (7. ξ2 . Numerical integration of strain matrices (a) Compute φ = [1. MIXED FINITE ELEMENT METHODS 55 ¯ σ stresses.82) gI = 2.85) (7.79) where ¯ ¯ ¯ ¯ ¯ KIJ = (Kdev )IJ + kI bJ + bT kT + bT kvol bJ I J I (7.78) ˆ ˆ Using the solutions to 7.72 and 7.

92) (Kdev )IJ = kI = Ωe BT d φ dΩ I φT dvol φ dΩ Ωe kvol = 6. Mixed Pressure ¯ (a) Compute p = φ(ξ) h−1 π 5. Residual and Stiﬀness Integrals (7.e.87) (7..93) 7.95) . MIXED FINITE ELEMENT METHODS 1 T ¯ m σ 3 φT p dΩ ¯ Ωe 56 p = ¯ ¯ π = 4.91) (7. p.4 Hu-Washizu Variational Theorem: General Problems The ﬁnite element approximation for the mixed formulation of a general linear elastic material (i.89) (7.CHAPTER 7. θ) = 1 2 + p( T Ωe D dΩ − Ωe T D 0 dΩ (7.90) (7. 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 (7.94) · 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. anistotropic behavior) may be written for a typical element as Πe (u.

CHAPTER 7.98) (7.97) (7. 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 (7.103) The mixed variational terms become Πe (u. MIXED FINITE ELEMENT METHODS which when introduced into the variational theorem gives Πe (u.100) (7.99) (7. 7. θ) = + 1 2 Idev BI uI + Ωe 57 1 m φ(ξ) θ 3 T D Idev BI uI T I 1 m φ(ξ) θ dΩ − 3 Ωe Ωe 1 Idev BI u + m φ(ξ) θ 3 φT φ dΩ θ Ωe D 0 dΩ (7.105) .96) + pT φT bJ dΩ uJ − For symmetric D. 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. ΘT gJ dη (KT )J −h k θ co 0 (Pdev )I − 0 π0 (7.101) (7.104) The ﬁrst variation of Eq. p.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. ΠT .104 may be written in the matrix form J (Kdev )IJ gI (Kco )I u dΠe T 0 −h p = (UI )T . p.

CHAPTER 7. and the variation in the volume change.7.108) (7. .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. this requires a computation of two d terms as 1 Dm 3 (7. and.105 may be solved at the element level to give the parameters for the volume change. 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 (7.113) (7.112) 1 1 T m D m = mT d 9 3 (7. thus. Essentially. MIXED FINITE ELEMENT METHODS 58 Recall that the terms which multiply the variations in pressure. Π.110) I co I This operation may be performed after all the integrals over the element are evaluated. the second row of Eq. are associated with individual elements.7.111) and using these in the remaining equations instead of d (note. 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.115) 3 If D is not symmetric. equations Eq. Θ.107) Thus.114) dR = D m . deﬁne dvol = Then D Idev = D − d mT or Idev D = D − m dT (7. as θ = h−1 gJ uJ (7. dL = DT m (7.112 through Eq. θ.109) (7.117 must be modiﬁed. we let3 d = Also. Accordingly. 7.

123) (Kco )J = and k = Ωe dvol φT φ dΩ (7. .CHAPTER 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 (7.124) The matrix for the initial strains is computed as π0 = Ωe φT dT 0 dΩ (7.116) Finally.125) which is a 1 × 6 vector.122) (7. MIXED FINITE ELEMENT METHODS which gives 1 1 Idev D m = (D − m dT ) m = d − dvol m = ddev 3 3 59 (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 (7.118) (7.117) (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.119) (7.120) Ddev (7.

127) This element is often called B1-P0 (order 1 interpolations for the brick element. The other matrices in the stiﬀness are h = h = Ωe dΩ = Ωe (7.133) and gI = Ωe bI dΩ (7. this matrix is zero. for the order 1 quadrilateral. In 2-dimensions the element is Q1-P0. for example. it is possible to use the interpolations φ(ξ) = 1 x1 (ξ) x2 (ξ) (7.130) which is a column vector (of size 1 × 24). the Q2-P1 element uses quadratic interpolation for displacements (Lagrange interpolations) and linear for the pressure with φ(ξ) = 1 ξ1 ξ2 (7. order 0 for the pressure/volume change).129) The matrices for the B1-P0 (or Q1-P0) element reduce to (Kco )J = Ωe dT BJ dΩ dev (7. For isotropy kvol is the bulk modulus times the element volume.128) Alternatively. For isotropy.1 Example: Interpolations linear for u and constant φ As an example.4. Higher order elements are also deﬁned.134) .131) k = Ωe which is a 1 × 1 matrix and for constant dvol becomes k = kvol = dvol Ωe (7. MIXED FINITE ELEMENT METHODS 60 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 (7.132) where Ωe is the volume of the element.CHAPTER 7.126) where ξiI are the values of the natural coordinates at the I-node. The volume stiﬀness becomes: dvol dΩ (7. The interpolation for the pressure (and volume change) is constant φ1 = 1 (7.

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

(s) u. ˜. and written as (u. ˜) = (s) u + ˜(ξ) (8.1) The strain tensor is expressed as an additive sum of the symmetric gradient of the displacement vector. σ. and the enhanced strains. In the enhanced strain method we again use the Hu-Washizu variational principle.Chapter 8 Enhanced Strain Mixed Method 8.2) 62 . the method does not have the deﬁciencies which are present in the earlier works. which we recall may be written for linear elasticity as Π(u. Enhanced strains provide great ﬂexibility in designing accurate ﬁnite element models for problems which have constraints or other similar types of diﬃculties. = + 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. however.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]. The enhanced strain method is related to earlier works which utilized incompatible displacement modes.

4) (8. the enhanced strains must vanish. We note Eq.8) 2. σ. ˜) = − Ω 1 2 ( Ω (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ω ( (s) u + ˜)T D dΩ − (8. the following Euler equations are obtained for the domain Ω: 1. ENHANCED STRAIN MIXED METHOD 63 If we again ignore the integral for the body force and the traction and displacement boundary integrals. 8.10) In addition the boundary conditions for Γu and Γt are obtained. upon use of Eq. from 8.3) 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 (8.5) (8. respectively. at the solution. While the enhanced strains vanish pointwise at a solution.8 and 8. 8. Substitution of this result into the remaining equations yields the appropriate displacement equations of equilibrium and constitutive equation for linear elasticity.9 implies that.2 the remaining terms become Π(u.10. in an approximate .6) ( Ω (s) U)T D ( (s) (s) u + ˜ − 0 0 ) dΩ ST ˜ dΩ Ω ˜ ET [D ( u + ˜ − ) − σ] dΩ − (8. Balance of momentum div [D ( (s) u + ˜ − 0 )] + bv = 0 (8.CHAPTER 8. Constitutive equations D( (s) (8. Strain-displacement equations on the enhanced modes ˜ = 0 3.9) u + ˜ − 0 ) − σ = 0 (8.7) After integration by parts of the variation of the displacement gradient term (and also considering the body force term).

stresses.CHAPTER 8. From Eq. ENHANCED STRAIN MIXED METHOD 64 scheme based upon the enhanced strain method this is not the case.11) and σ(ξ) = φα (ξ) σ α (t) (8. 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. ) (8. In the enhanced strain principle.16) Substituting the approximations for displacements. the strain approximations are now given by (u. σ. Using the interpolations described above.18) . σ. ) ≈ e Πe (u. Π(u. as will be shown below.17) η σ α = σ α + η Sα η and ˜ ˜α = ˜α + η Eα η (8. just as the balance of momentum and constitutive equations are approximated by ﬁnite element solutions. ˜) = (s) u + ˜(ξ) (8. The enhanced strains will only vanish in some integral sense over each element. Appropriate interpolations for displacements and stresses are the same as given previously for each element.15) The variational expression in each element becomes Πe (u.12) respectively. σ. 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. σ. displacements appear up to ﬁrst derivatives. and enhanced strains and replacing with uI = uI + η UI (8. ˜) = − Ωe 1 2 ( Ωe (s) u + ˜)T D ( 0 (s) u + ˜) dΩ σ T ˜ dΩ Ωe ( (s) u + ˜)T D dΩ − (8. Accordingly. ) = Πh (u. Accordingly. and are thus u(ξ) = NI (ξ) uI (t) (8. while the stresses and enhanced strains appear without any derivatives. 8.19) (8.2.14) It should be noted that diﬀerent interpolations are introduced for the stress and the enhanced strain terms.13) where the approximations for the enhanced strains are assumed as ˜(ξ) = ψ α (ξ) ˜α (t) (8.

8.20) (8. ENHANCED STRAIN MIXED METHOD gives the ﬁrst variation in each element as dΠe ˜ = (UI )T . The interpolations for φα are orthogonal to the interpolations ψ β . This is not perfect since we will not obtain a method to compute the σ β directly from the variational formulation. the variational equations in each element reduce to dΠe = dη ˜ (UI )T .28) which is the solution to be followed here.29) . 3. which is not a useful result. The β are zero.28.CHAPTER 8. 8.20 is given in each element by Qαβ β = 0 (8.22) (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.23) (8. (Eα )T . (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. which means that Qαβ = 0 (8. (Eα )T ˜ KIJ ΓβI T ˜ ˜ ΓαJ Hαβ uJ ˜β − P0 I ˜α π0 (8.24) (8.25) (8. A combination of options (a) and (b).27) There are at least three possible ways this may be used: 1. For a formulation which satisﬁes Eq.21) (8. 2.

ENHANCED STRAIN MIXED METHOD 66 Since the interpolations for the enhanced strains are assumed for each element independently.29 may be solved at the element level giving ˜ ˜β = (Hαβ )−1 ˜ ˜α π 0 − ΓαJ uJ (8. for outputs and stress projections to nodes.31) (8.CHAPTER 8. the variation in each element may be written dΠe dη = η=0 Ωe ( (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8.35) It is noted that the orthogonality condition ST ˜ dΩ = 0 Ωe (8. Simo and Rifai suggest using a least square projection technique to obtain the stresses.37) . the second of Eq. Thus. however. 8. 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. the residual in each element becomes: RI = FI − Ωe BT σ dΩ I ˜ (8. With the above description.32) (8. the stresses which are directly utilized in the variational equation Eq.33) 8.34) In subsequent development we shall use these stresses for all calculations of arrays. as well as.30) which may be substituted into the ﬁrst equation to give dΠe ˜ ˜ = (UI )T KIJ uJ − P0 I dη where and ˜ ˜T ˜ ˜ KIJ = KIJ − ΓβI (Hαβ )−1 ΓαJ ˜ ˜T ˜ ˜α P0 = P0 − ΓβI (Hαβ )−1 π 0 I I (8.36) has been incorporated in the above variation.7 may be deduced as ˜ σ = D( (s) u + ˜ − 0 ) (8.

39) ˜ which represents a transformation of the local enhanced strains.42) (8.44) j0 −1 ˜ F E(ξ) j(ξ) 0 j0 T F Σ(ξ) j(ξ) 0 (8.38) Note that the residual for the enhanced modes will vanish at a solution since it belongs to a single element. 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 ˜ Rα = − Ωe ψ T σ dΩ α ˜ (8. Accordingly. and gradients of local coordinates.41) (8. J0 .CHAPTER 8. Σ. σ. using the transformation deﬁned at the element center. E. 8. ξ.40) The transformations may also be written in matrix form as ˜ = σ = where for 2-dimensional problems ˜ ET = ˜ ˜ ˜ ˜ E11 E22 E33 2 E12 (8. The weighting by the jacobian determinant terms is motivated by the gradient of the shape functions. it is useful to express the enhanced strains in a similar form. expressed on the bi-unit square to the global strains. Using tensor notation we introduce the representations ˜ = j0 −T ˜ J0 E(ξ) J−1 0 j(ξ) (8. on the bi-unit square element to the global stresses. a transformation of the local stresses.3 Construction of Enhanced Modes The construction of the enhanced modes depends crucially on the orthogonality requirement being satisﬁed for each element. Similarly. is given by σ = J0 Σ(ξ) JT 0 These transformations have the property that tr(σ ˜) = j0 ˜ tr(Σ E) j(ξ) (8. ˜. the constant part of the jacobian matrix. 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.43) . j(ξ).

for non-linear materials.48) Thus.45) and ˜ and σ have similar ordering. It should be noted however. A number of alternatives are discussed in the paper by Simo and Rifai [16]. 8. the satisfaction of the orthogonality condition may be accomplished by constructing the interpolations in the natural coordinate system and transforming to the global frame using Eq. 8.49 and Eqs. Here we consider the simplest form.50 contain only linear polynomials in ξ and.CHAPTER 8.48.50) for the enhanced strains are used. 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.43.49) ξ1 0 0 0 0 ξ2 0 0 ˜ E = 0 0 0 0 0 0 ξ1 ξ2 (8. contrary to what is necessary when using incompatible modes.47) ˜ ΣT E d2 = 0 2 (8. These interpolations have been incorporated into the element routine elmt04 which has been developed for a linear elastic-viscoelastic material.42 and Eq.46) j0 ˜ ΣT E j(ξ) (8. 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. 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. For the simplest form.41 may be written as σT ˜ = The integral over the element becomes σ T ˜ dΩ = j0 Ωe (8. satisfy the orthogonality condition Eq. which indeed is identical to the modiﬁed incompatible mode formulation [22]. 8. thus. that no ad-hoc assumptions are required in the enhanced formulation. the interpolations given by Eqs. 8.51) Thus. ENHANCED STRAIN MIXED METHOD ΣT = Σ11 Σ22 Σ33 Σ12 68 (8. . as well as. 8.

for outputs and stress projections to nodes.35 to 8. 8. A popular scheme is Newton’s method. W ( ).53) We note that for the linear material model discussed previously that W( ) = 1 2 T D − T D 0 (8.CHAPTER 8. through ∂W σ = (8.57) I ˜ Ωe Similarly.4 Non-Linear Elasticity For a non-linear. which may be summarized as: . where σij = ∂W ∂ ij (8. the residual in each element becomes: RI = FI − BT σ dΩ (8. Thus. as well as. the residual for the enhanced modes is computed from ˜ Rα = − Ωe ψ T σ dΩ = 0 α ˜ (8.58) ˜ We note above that at a solution the residual. 8.5 Solution Strategy: Newton’s Method The solution to a non-linear problem is commonly computed using a sequence of linear approximations. hyperelastic material the stresses are computed from a strain energy density function. Rα .52) ∂ The partial derivative is understood in terms of components.56) In a manner identical to the linear elastic material.38) dΠe = dη ( Ωe (s) ˜ U)T σ dΩ + Ωe ˜ ˜ ET σ dΩ (8.55) = (s) u + ˜ In subsequent development we shall use these stresses for all calculations of arrays. should vanish independently in each element.54) For the enhanced formulation the computation of stresses is given by ˜ σ = ∂W ∂ (8. for the enhanced formulation the variation in each element may be written as (see Eqs. ENHANCED STRAIN MIXED METHOD 69 8.

8.61) x=x(i) (8.62) 4. 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. F(i) . is the Jacobian or tangent matrix for the equations. Using Newton’s method on the set of equations deﬁned by Eq. Convergence may be assessed from | dx(i+1) | < tol | x(i+1) | where | x | is the length of the vector. ENHANCED STRAIN MIXED METHOD 1. 3. 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. Repeat steps b.64) In the above. F(i) = ∂f ∂x ∂f ∂x dx(i+1) = 0 x=x(i) 70 (8. Solve the linear problem dx(i+1) = − (F(i) )−1 f (i) and update the solution as x(i+1) = x(i) + dx(i+1) In the above.65) which expands to (i) KIJ = Ωe ˜ ∂σ BI ∂ (i) ∂ dΩ = ∂uJ ˜ (i) BI Dt BJ dΩ Ωe (8. Given the set of equations f (x) = 0 where x are the dependent variables.60) (8.66) .CHAPTER 8.58 above gives the problem RI ˜ (i+1) Rα (i+1) (8.57 and Eq.) until the solution converges to within a tolerance.59) (8. x. 2. tol.) and c. the terms in the Jacobian are deﬁned as (i) KIJ ∂RI = − ∂uJ (i) (8. .

the assembled equations become ˜ (i) KIJ duJ(i+1) = e e ˜ (i) RI + F I (8. Thus after adding any nodal forces.64 is complete at the element level.70) Since the second equation in Eq.69) and ˜ (i) Hαβ = (8. 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. 8.68) ˜ (i) ψ T Dt BJ dΩ α ˜ (i) ψ T Dt ψ β dΩ α Ωe (8. FI .67) deﬁne the tangent moduli for the material.76) The incremental displacements also may be substituted back into Eq. Accordingly. ENHANCED STRAIN MIXED METHOD where ˜ ∂σ ˜ (i) Dt = ∂ 71 (i) (8.75) which may be solved for the incremental nodal displacements. duJ(i+1) .77) . ˜ (i) ΓαJ = Ωe ∂2W ∂ ∂ (i) (8.72) (8. For the non-linear elastic material ˜ (i) Dt = Similarly.73) (8.CHAPTER 8. we may perform a partial solution by static condensation. these may then be used for the update ˜β(i+1) = ˜β(i) + d˜β(i+1) (8. After the solve.71 to compute the increments to the enhanced modes.74) The reduced ﬁrst equations may be assembled into the global equations.71) (8. the new nodal displacements are updated uJ(i+1) = uJ(i) + duJ(i+1) (8.

81) (s) u(i) + ˜(i. for this algorithm.71 for the later update of the enhanced modes. ˜ 2. as well as use for subsequent steps for the global i-iterations or to compute stresses.CHAPTER 8. Update the solution ˜β(i.k) Hαβ = Ωe (8.80) with ˜ ˜ (i.k) ˜ d˜β(i. Solve for the increment ˜ (i.k) + d˜β(i.k+1) (8.k) = ˜β(i−1) (8. 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.k+1) = (Hαβ )−1 R(i.k) 3. ˜β(i) . Consequently. (8. Set ˜β(i) = ˜β(i.k+1) = 0 where now ˜ (i. Compute the linear part of Rα as ˜ ˜ (i.k) ) − Hαβ d˜β(i. 8. until convergence achieved (or a set number of k-steps is completed).k) Rα (uJ (i). Set k ← k + 1 and repeat Steps 2. it is necessary to save the arrays used in Eq.k+1) (8. to 4. In the enhanced element for 2-dimensional plane strain applications in FEAP. J Accordingly. 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) = ∂ σ Dt ∂ (8. the arrays are moved into history arrays using a pmove routine. 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.84) and save for the next global iteration.82) . 1.k) α 4. For k = 0 set ˜β(i. 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. ˜ The alternate algorithm is given by linearizing the residual.k+1) = ˜β(i.78) where a single superscript i denotes the value of ˜β computed in the last global iteration.83) 5. Rα . 6.79) ˜ (i.k) ψ T Dt ψ β dΩ α (8. ˜β(i.

6) ul(ndf.1) ul(ndf.2: Element Local Arrays The only information to be stored is the ˜β(i) . ∆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. The deﬁnitions of the entries in the local array are given in Table 8. 8. J.nen.7) Description local uJ (i) local ∆uJ (i) local duJ (i) ˙ local uJ (i) ˙ local uJ (i) ˙ local uJ (i − 1) used for b.4) ul(ndf.nen. .72 to Eq. however.nen. using only 2 or 3 iterations generally suﬃces (even though convergence may not be achieved for the ﬁrst few values of the i-global iterations). In FEAP. 8. leading to Eq. ˜ (i) ˜ σ (i) .2. 8. The global arrays are passed to each element in a local array.k) putation of Rα and Hαβ .3) ul(ndf. and the tangent moduli.1: Element Local Arrays Array ul(ndf.75 thus simplifying slightly the steps involved.75 for the global steps.nen.k) (i. Dt .2) ul(ndf. linearization with respect to both uJ and ˜B is performed. and a class of viscoplastic materials. elasto-plastic materials. 8.1.CHAPTER 8. In subsequent presentations we shall discuss the construction of these steps for linear viscoelastic materials.i). on uJ (i) Problems All All All Transient Transient Transient All1 73 Table 8.nen.c.nen.5) ul(ndf.nen. ENHANCED STRAIN MIXED METHOD Vector Deﬁnition uJ (i) Description Current solution value at each node.nen. Once the k-iteration is completed. If the k (i) iteration is converged.72 to Eq. The array ul contains information for the current element according to the deﬁnitions in Table 8. While the above process has been illustrated for the non-linear elastic material. the uJ (i) nodal displacement vector and the ∆uJ (i) and duJ (i) nodal incremental vectors are retained in global arrays. it may be directly extended to any material for which we can iteratively compute the stresses. The algorithm requires repeated com(i. ul(ndf. the Rα is zero in Eq.

4) = e + In the presentation given here we assume that the pressure-volume parts of the behavior are governed by a linear elastic model p = Kθ (9. and θ is the volume change deﬁned in matrix form as θ = mT (9.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. s is the stress deviator.6) K = 3 (1 − 2 ν) 74 . and p is the mean (pressure) stress deﬁned in matrix form as 1 p = mT σ (9.3) 3 is strain. in matrix form the stress and strain may be split as σ = s + mp and (9.5) where K is the bulk elastic modulus deﬁned in terms of Young’s modulus and Poisson’s ratio as E (9. e is the strain deviator. In the discussion to be presented here we assume the material is linear and isotropic.1) 1 mθ (9.2) 3 where σ is the Cauchy stress. Accordingly.

12) This form of the representation is equivalent to a generalized Maxwell model (a set of Maxwell models in parallel). and the stress response. The relaxation modulus function is deﬁned in terms of an idealized experiment in which.11) ˙ qi + 1 i ˙ q = e λi (9.7) 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. e.CHAPTER 9. 75 Linear viscoelastic behavior may be stated in the form of diﬀerential equation models or in the form of integral equations. qi . Alternatively. In the diﬀerential equation model the constitutive equation may be written as P (s) = 2 G Q(e) (9. LINEAR VISCOELASTICITY The deviatoric parts are assumed to satisfy a linear viscoelastic model. An alternative to the linear viscoelastic model in diﬀerential form is to use an integral equation form. at time labeled zero (t = 0). h qi (t) = C exp h and variation of parameters on C to give t −t λi (9.10) is identical to the elastic shear modulus. The set of ﬁrst order diﬀerential equations may be integrated for speciﬁed strains.8) (9. The integral form equivalent to the above is expressed in terms of the relaxation modulus function. e0 .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.9) (9. a specimen is subjected to a constant strain. s(t). is measured. the operator may be written as N s = 2 G (µ0 e + i=1 µi qi ) (9. For a linear material .13) qi (t) = −∞ exp − t−τ ˙ e(τ ) dτ λ1 (9. The integral for each term is given by the homogeneous diﬀerential equation solution.

In the subsequent discussion we will consider the generalized Maxwell material and let N be 1 (i.22) . 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.CHAPTER 9.. Accordingly.16) It is noted that the above form is a generalization of the Maxwell material. This relation may be written as s(t) = 2 G(t) e0 (9.20) λ1 where µ0 + µ1 = 1 (9.17) or the alternate form G(t) = G (µ0 + N µi exp i=1 −t ) λi (9. it is usually assumed that the material is undisturbed until a time identiﬁed as zero.19) With this form the integral equation form is identical to the diﬀerential equation model for the generalized Maxwell material. 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. 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 (9.21) In applications involving a linear viscoelastic model. −t G(t) = G (µ0 + µ1 exp ) (9.18) where µ0 + N µi = 1 i=1 (9. LINEAR VISCOELASTICITY 76 a unique relation is obtained which is independent of the magnitude of the applied strain.e. the standard linear solid).15) where G(t) is deﬁned as the shear relaxation modulus function. At time zero a strain may be suddenly applied and then varied over subsequent time. The addition of more terms may be easily accommodated based upon the one term representation.

9. 9. tn ≤ τ ≤ t (9.26) the above separation gives t i (t) = i (tn ) + tn 1 1 exp τ ˙ e(τ ) dτ λ1 (9.23) where subsequently the 0 limit on the integral is understood as 0+ .32) . and the last term covers the subsequent history of strain. The result of this separation when applied to Eq. the second term includes a jump term associated with e0 at time zero.31) ∆t where en denotes the value of the strain at time tn and ∆t denotes the time increment t − tn . Accordingly.20 into Eq. 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 (9.29) −t t τ ˙ exp e(τ ) dτ (9.23 gives s(t) = 2 G [µ0 e(t) + µ1 exp −t (e0 + λ1 t exp 0 t ˙ e(τ ) dτ )] λ1 (9. we divide the integral as t tn t (·) dτ = 0 0 (·) dτ + tn (·) dτ (9. thus ∆h = exp ˙ e(τ ) ≈ e(t) − en . Substitution of Eq. 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 (9.16 gives t s(t) = 2 G(t) e0 + 2 0 ˙ G(t − τ ) e(τ ) dτ (9.30) λ1 tn λ1 The strain rate is now approximated as constant over each time increment tn to t. 9.CHAPTER 9.24) It remains to evaluate the integral.28) −∆t 1 h + ∆h λ1 n (9.27) Including the negative exponential multiplier term gives h1 = exp and then h1 (t) = exp where −t 1 i λ1 (9.

we need to compute ∂s ∂ = ∂s ∂e ∂e ∂ = ∂s Idev ∂e (9.35) This form requires only one evaluation of an exponential term.34) h1 (t) = exp − ∆t − ∆t 1 [exp h + (e − en )] 2 λ1 2 λ1 n (9. Thus. Thus.37) The inclusion of more terms in the series reduces to evaluation of additional hi (t) integral recursions. The result gives ∆h = λ1 ∆t 1 − exp −∆t λ1 (e − en ) (9.38) . While the above form is easy to evaluate it has problems when the size of the time step is changed. Thus. however. 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. The constitutive equation now has the simple form s(t) = 2 G [µ0 e(t) + µ1 h1 (t)] (9. For very small time steps a series expansion may be used to yield accurate values.36) This approximation produces a singular ratio for zero time steps.33) where tn+ 1 denotes the time 1 (tn + t). a very large value of the time step producing a zero value. It may also be extended for use with thermorheologically simple materials. a more stable form is used in FEAP where the integral over the time step is evaluated in closed form [23]. The implementation of the viscoelastic model into a Newton solution process requires the computation of the tangent tensor.CHAPTER 9. Accordingly. LINEAR VISCOELASTICITY The recursion then becomes t 78 exp tn tn+ 1 τ 2 dτ ≈ ∆t exp λ1 λ1 (9. the limit value is well behaved at a unit value. a zero value of the time step produces a correct answer. This form gives a recursion which is stable for small and large time steps and gives smooth transitions under variable time steps. the form is doubly asymptotically accurate. The use of ﬁnite diﬀerence approximations on the diﬀerential equation form directly does not produce this property. as well as. Furthermore. 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.

9.CHAPTER 9. 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. Since the material is linear. Again we note that for zero ∆t the full elastic modulus is recovered.40) ∂h1 = 2 G [µ0 I + µ1 ] ∂e (9. use of the consistently derived tangent modulus terms leads to convergence in one iteration (the second iteration produces a zero residual). 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.43) for the elastic shear modulus.39 becomes ∂h1 ∂e Thus.41) (9. LINEAR VISCOELASTICITY 79 where Idev is the deviatoric operator identiﬁed previously. .39) The only modiﬁcation from a linear elastic material is the substitution of the factor Gvisc = G [µ0 + µ1 ∆h1 (∆t)] (9.42) = ∆h1 (∆t). I (9. the tangent tensor is given by ∂s = 2 G [µ0 + µ1 ∆h1 (∆t)] Idev ∂ (9. 0 The above formulation is incorporated into the subroutine viscoe.

A back stress is deﬁned as α which is related to the plastic 80 .4) where f is a loading function and γ is a scalar rate term called the plastic consistency ˙ parameter. If the material is nonlinear hyper-elastic we may deduce the stress from the expression for the elastic strain energy as ∂W σ = (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. The plastic consistency parameter. is zero for elastic behavior and pos˙ itive for plastic behavior. γ. The inelastic component of the strain rate is related to the gradient of a loading function with respect to stress. ˙p = γ ˙ ∂f ∂σ (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. Accordingly.Chapter 10 Plasticity Type Formulations 10. 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.

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

For an inelastic step the stresses must satisfy (1. for the present we restrict our attention to plasticity as deﬁned by Eq. and ep .12) n+1 = n αT R = αn n+1 epT R = ep ¯n+1 ¯n and γn+1 = 0 ˙ TR Thus for linear elasticity σT R = D ( n+1 n+1 (10.16) The trial stress given by Eq. integrating non-linear terms using a backward Euler implicit method between tn and tn+1 .15) pT R n+1 ) − (10.11 to determine if the step is elastic or whether inelastic terms should be active. the plastic strain is given by p n+1 = p n + λn+1 ∂f ∂σ (10. 4] the algorithm may be extended to include a variety of modeling concepts for the limit behavior. Accordingly.17) n+1 and the back stress by ∂f p ˆ αp n+1 = αn + Hkin λn+1 ∂σ (10. To initiate the solution at tn+1 a trial state is computed assuming the step is entirely elastic. PLASTICITY TYPE FORMULATIONS 82 10. 10.11 above. This implies that there will be no ˙ change to p .6 and Eq.2) for the time tn+1 which requires the rate equations for p and α to be integrated over the time increment. a discrete solution at time tn is deﬁned in terms of the state σ n .18) n+1 . 3.CHAPTER 10. or ep during an elastic increment. The step is initiated by taking the ¯ trial values for plastic quantities pT R p (10. however.16 is checked in Eq. 10. 10. If the state at tn+1 is elastic the stresses (as well as other state variables) are set equal to the trial value. 18. n+1 .13) (10. p . Accordingly. A very eﬀective method to integrate plasticity equations is the operator split method with a return map concept [17. a correction is required to include the inelastic terms. 10. αn .2 Solution Algorithm for the Constitutive Equations The solution of the above set of equations may be eﬀected numerically using a variety of algorithms. otherwise. α. Recall that a step is elastic when γ is zero. The ¯n n solution is then advanced to time tn+1 by specifying the strain.14) (10.

18 and 10.20 constitute a set of non-linear equations in terms of the values of σ n+1 . α(j) .20 as residual equations.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 . 10. The iteration counter is shown as a superscript (j) and initial iterate values are taken as the trial stress and zero λ( 0).21) n+1 (10.3.22) ∂f ∂f ∂Σ ∂f = = − (10. 10. Before proceeding with Newton’s method we note that the following relations hold ∂f ∂f ∂Σ ∂f = = ∂σ ∂Σ ∂σ ∂Σ and (10. αn+1 and λn+1 which must be solved for each stress point and each time step of interest. normally. The iterative solution is continued until the norm the residuals are within acceptable tolerance values (e. 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.18 and Eq. To simplify the notation the subscripts on n + 1 are omitted.g.19) (10.26) we may linearize the equations to obtain (note the iteration counter j is omitted in the coeﬃcient array for simplicity) (j) ∂2 ∂f ∂ ˆ (λ ∂Σ ) C −λ ∂Σf2 Rσ dσ (j+1) ∂λ (j) ˆ ∂f ∂2f ∂2f ∂ (j+1) ˆ ˆ = Tα (10. A Newton method may be used to solve the equations. 10. ep (j)) ¯ (10.4 produces ¯n ep ¯n+1 = ep + λn+1 ( 3 ∂f ∂f 1 · )2 2 ∂σ ∂σ (10.11 at tn+1 gives gn+1 = fn+1 = 0 Finally. treating the equations Eq.25) α ∂Σ and (j) Rf = − f (σ (j) . 10.CHAPTER 10.23) ∂α ∂Σ ∂α ∂Σ Thus.20) The set of equations Eqs. integration of Eq. in the form ∂f R(j) = − p − D−1 σ (j) − λ (10. 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)..24) σ n ∂Σ ∂f ˆ R(j) = αn + Hkin λ − α(j) (10. 10. evaluating Eq.

however. the residuals will vanish. That is we need to perform a new solution to see if the strains we used to compute the stresses are correct.34) At convergence for the given strain. by solving KIJ uJ = FI + RI (10.35) 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. as before. The solution is terminated whenever the norms of the residuals are smaller than a selected small tolerance. Rα = 0 . The linearization of the residuals with respect to an increment of strain yields Rσ = d . PLASTICITY TYPE FORMULATIONS where A = and 84 ∂Y ∂¯p e p ∂λ ∂¯ e (10. respectively.36) ˆ ∂Σ ∂λ ∂Σ ∂Σ T T ∂f ˆ ˆ ˆ D31 D32 D33 − ∂f −A ∂Σ ∂Σ . if we now consider a linearization with respect to strain only Rσ contributes to the change. In addition it is necessary to compute the tangent moduli. 10. σ (j+1) = σ (j) + dσ (j+1) (10.33) e e (10. R f = 0 (10.29) The solutions to Eq. 10.30) α(j+1) = α(j) + dα(j+1) and λ(j+1) = λ(j) + dλ(j+1) (10.CHAPTER 10. Once convergence is achieved for each stress point evaluation (i. the stresses may be used in the ﬁnite element to compute each element residual. Accordingly.28) ∂2f ˆ C = D−1 + λ ∂Σ2 (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. Dt . for use in the element stiﬀness matrix (if one is used) for the next iteration on the momentum balance equation.27 are computed and added to obtain the next iterates.32) deﬁne the next iterates. This is accomplished. .31) where KIJ and RI are the element stiﬀness and residual..e. to compute the stress at each Gauss point for a given strain). 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.

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. [13].CHAPTER 10.41) . 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. ep ) ≤ 0 ¯ (10. Thus.42) For our study on inelastic behavior. The method provides the best estimates for the parameters and their sensitivities to errors or inconsistencies in the data. θ. Multiple yield surfaces may be included by modifying Eq. al. and volumetric (spherical).43) and θ = θe + θp (10.4 to K ˙p = k=1 γk ˙ ∂fk ∂σ (10. 10. the tangent moduli for the next momentum iteration is ˆ Dt = D11 85 (10.3 Isotropic plasticity: J2 Model As in previous developments.40) An active condition for each surface is denoted when γk ≥ 0. 10.44) 1 mθ 3 (10.39) with each part of the yield surface described by a separate function gk = fk (σ. et. the decomposition into elastic and plastic parts may now be expressed as e = ee + ep (10. the strain is again split into deviatoric. parts and expressed in matrix notation as = e + where θ = mT (10. α.38) Except for giving the form of f this completes the speciﬁcation of the general algorithm. e.37) (10.

includes the eﬀects of linear isotropic hardening. the limit equation as g = f (s. PLASTICITY TYPE FORMULATIONS The stress also is split into the deviatoric. s. Thus θp vanishes and the yield function can depend only on the deviatoric part of the stress.46) 3 With the above splits the isotropic linear elastic constitutive equations are given by p = p = K (θ − θp ) and s = 2 G (e − ep ) (10.49) J2 = and J3 = det (s) (10.50) (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. The invariants of s are denoted as J1 . J2 . through 2 Y = (σy + Hiso ep ) ¯ (10. We write this model using 1 2 J2 = (sT s) 2 = s (10. and pressure (spherical) parts as σ = s + mp where 86 (10.45) 1 T m σ (10.55) A simple calculation shows that ∂f ∂f = Idev ∂σ ∂s . thus. For isotropic materials the yield and loading function may be expressed in terms of the invariants of stress and back stress.CHAPTER 10. ∂f ∂f ∂Σ ∂f = = − (10. and J3 and given by J1 = m T s = 0 (10. α.53) 1 T s s 2 (10.48) In the developments below we restrict plasticity to the deviatoric part only. ep ) = ¯ s − α − Y ep ≤ 0 ¯ (10.56) ∂α ∂Σ ∂α ∂Σ .51) The simplest formulation is where the function depends only on J2 . σy . The back stress adjusted value Σ is given by Σ = s − α (10.54) 3 and. ∂f ∂f ∂Σ ∂f = = ∂s ∂Σ ∂s ∂Σ .

the evolution of the back stress satisﬁes ˙ α = 2 ∂f 2 Hkin γ ˙ = Hkin γ n ˙ 3 ∂σ 3 (10. thus. our model is completed by giving the evolution equations for plastic strain and eﬀective plastic strain in the form ˙ ep = γ n ˙ t (10. mT n = 0 . nT n = 1 (10.66) (10.60) (10.63) (10.69) αn+1 = αn + 2 Hkin λn+1 nn+1 3 2 λn+1 3 − Yn+1 ≤ 0 ep ¯n+1 = ep + ¯n gn+1 = Σn+1 Σn+1 = sn+1 − αn+1 and Yn+1 = 2 2 (σy + Hiso ep ) = Yn + Hiso λn+1 ¯n+1 3 3 (10. we can conclude that the back stress evolves such that it is a purely deviatoric quantity.57) Thus. Thus.70) .67) (10.64) (10. mT α = 0 With this fact we then have the following important properties mT Σ = 0 . PLASTICITY TYPE FORMULATIONS (where we recall that Idev = I − ∂f = ∂Σ 1 3 87 m mT ).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 (10.58) Noting that at the initial state α is zero.CHAPTER 10. and Σ Σ = n = ∂f ∂σ (10.59) Based upon the above all aspects of the J2 plasticity model are restricted to deviatoric components only.61) (10.65) (10.68) (10.

10. n+1 epT R = ep n+1 n .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 (10.71) 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. αT R = αn n+1 . The second part of the algorithm solves the discrete rate equations using the trial values as initial conditions. 10.74 gives also that sT R − αn is in the n+1 direction nn+1 . 10.CHAPTER 10. 10.75 with Eq.75) that is.72 in Eq. Accordingly. and may be written as the scalar equation Σn+1 − ΣT R n+1 + 2 (G + 1 Hkin ) λn+1 nn+1 = 0 3 (10.73) Noting that Σn+1 = Σn+1 nn+1 Eq. The solution is performed using a trial state based upon the assumption that λT R is zero. using Eqs.78) .66 from Eq. If the trial values violate the limit equation.64 gives sn+1 = sT R − 2 G λn+1 nn+1 n+1 Next subtracting Eq. it is necessary to perform the second part of the algorithm. 10.72) which may be used to check the limit equation. 10.65.70 yields the scalar equation.68 and Eq. If the limit equation is satisﬁed then the trial values deﬁne the solution at tn+1 . 10. the coeﬃcient must vanish to obtain a solution. Accordingly.73 gives Σn+1 = sn+1 − αn+1 = sT R − αn − 2 (G + n+1 1 Hkin ) λn+1 nn+1 3 (10. In the above ΣT R = sT R − αn n+1 n+1 Combining Eq.74) (10.79) nT R n+1 = ΣT R n+1 ΣT R n+1 (10. ΣT R − Yn = 2 [G + 1 (Hiso + Hkin )] λn+1 3 (10.71 and 10. epT R = ep ¯n+1 ¯n (10.77) (10. 10. 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 .

a check on satisfaction of the momentum equation must be made. In practice λn+1 is reduced slightly to that sn+1 is always slightly outside the limit yield condition. Similarly.82) Rα = α − αn − and R f = Yn + 2 Hiso λ − Σ = 0 (10.83) 3 In the above we have deleted speciﬁc reference to the values at tn+1 .77 will be nonlinear in λn+1 and a linearization and iteration process are required to obtain a solution. αn+1 . i. previously given as λn+1 . the linearization n becomes ∂n = ∂Σ 1 Σ (I − n nT ) = 1 Σ N (10.CHAPTER 10.80) 1 3 where tol is a small value. Writing appropriate residual equations as Rs = e − e p − n 1 s − λn = 0 2G 2 λ Hkin n = 0 3 (10. It should be understood that λ denotes the value of the solution in the tn+1 step. and the consistency parameter. If non-linear isotropic hardening is included or alternative forms for the limit equation are employed. 10. say 10−8 . 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. n+1 .78 is perturbed as λn+1 = ΣT R n+1 2 [G + − Yn (1 + tol) (Hiso + Hkin )] (10. 10. e. If the momentum equation is not satisﬁed for the current time. the solution to Eq. Accordingly.e. as shown above. PLASTICITY TYPE FORMULATIONS 89 as well as the plastic strain and the back stress. 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. We note that for the current strain. sn+1 . λn+1 . to avoid added complexity in the linearization performed below. tn+1 .84) . which we denote by de. the above equations are satisﬁed. if non-linear kinematic hardening is introduced the reduction to a scalar equation may also be complicated.. For the solution process developed here. The solution of the problem. Once the converged value for λn+1 is known and the stress state determined. the equation equivalent to Eq. and etc. this may be achieved by selecting as primary dependent variables the stress. the back stress. Accordingly. another iteration is necessary to improve the estimate for the state of strain. does not require a linearization or an iteration process.81) (10. however. for all the other variables.

. the linearization of Eqs.89) This result may be substituted into the third equation in Eq.90) Substituting this result back into the ﬁrst of equation Eq. Accordingly.87) where B and C are given by B = 2Gλ = Σ D 2Gλ ΣT R . This also permits the substitution of alternative limit equations without changing the solution to the ﬁrst part.CHAPTER 10. for the diﬀerential strains. and then substituting the result into the third equation to obtain a ﬁnal expression for dλ in terms of de.92) Finally.81 to 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 (10.87 yields the incremental equation which yields the tangent modulus matrix for the algorithm. 10. d . 10.91) G (Hiso + Hkin ) (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) 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. the tangent becomes ds = 2 G Idev − B (Idev − n nT ) − A n nT d 1 (10.88) and where we have noted that D = 1 + 2 (G + Hkin ) 3 λ Σ = ΣT R Σ (10. Thus.85 to obtain 2 G nT de = 2 [G + 1 (Hiso + Hkin )] dλ 3 (10. 10. PLASTICITY TYPE FORMULATIONS 90 Using this result. C = 2 Hkin λ 2 Hkin λ = 3 Σ D 3 ΣT R (10.93) See Appendix E for a discussion on the inverse of this type of matrix. we obtain ds = 2 G [I − B (I − n nT ) − A n nT ] de where A = G + 1 3 (10.83 yields the set of equations 1 λ I + Σ N − Σλ R N n T ds de 2G 2 λ Hkin 2 λ Hkin 2 0 (10.

through 2 (σy + Hiso ep ) ¯ (10. 10. ∆t) using a backward Euler evaluation of the integrals to obtain ∆t Φ [f (sn+1 .95) with Y the radius of the yield function which is related to a uniaxial yield stress. Other functional forms for Φ may be considered without conceptual diﬃculty. PLASTICITY TYPE FORMULATIONS 91 10. α.3.70. α. ep )] − ζ λn+1 = 0 ¯n+1 (10. 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].96) Y = 3 and.94 for the time interval tn to tn+1 (i.98) . the behavior is inelastic and the return map solution for the viscoplastic model is given by Eqs. here we restrict our attention to the simple case given by Φ (f ) = (f )m (10. the relationship for g becomes a constitutive equation describing the evolution for the consistency parameter. The function Φ together with the parameter ζ determine the rate dependency of the model. the stress state may lie outside the yield surface. 10.CHAPTER 10. For the viscoplastic model considered here. For trial stress values for which the yield function deﬁned by Eq. In viscoplasticity. ep ) = ¯ s − α − Y ep ¯ (10. we ˙ write g = Φ [f (s. the only modiﬁcation to the formulation is the replacement of the limit equation for g.94) where the yield condition. other models representing the problems of generalized plasticity and generalized viscoplasticity can also be developed by such replacement.97) where m is a positive integer power.e. αn+1 . 10. γ. 10. In this section we show how such a formulation may be easily extended to include rate eﬀects in the inelastic behavior. ep )] − ζ γ ≤ 0 ¯ ˙ (10.4 Isotropic viscoplasticity: J2 model The previous section presented the formulation and solution algorithm for a J2 classical plasticity model.95 exceeds zero. still is given by f (s.67. f . Indeed. thus. For viscoplasticity. Accordingly. For loading rates which are ﬁnite. Y and σy relate to the inelastic behavior in the limit as γ tends to zero (the rate independent ˙ limit). Perzyna considers alternatives for representing Φ.63 to 10. includes the eﬀects of linear isotropic hardening.. The formulation is completed by integration of the constitutive equation Eq. however. σy .69 and 10. All the other equations for the model remain as given in Section 10.

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. Accordingly. .101 to Eq.99) For the simple model used here for Φ. we obtain ∆t Φ [ ΣT R − Yn − 2 (G + 1 (Hiso + Hkin )) λn+1 ] − ζ λn+1 = 0 3 (10.75 with Eq. 10. 10. back stress. The extension to higher powers of m may be constructed using a Newton solution scheme to solve the non-linear scalar equation. the presence of ∆t in Eq. 10. PLASTICITY TYPE FORMULATIONS 92 The discrete consistency parameter may be obtained by combining the scalar coeﬃcient from Eq. and eﬀective plastic strain are computed using the same expressions as for the classical plasticity model.101) Comparing Eq. 10.98 to obtain a single nonlinear equation in λn+1 .100) In general. For nonzero ζ. 10.80 we can observe that the limit solution for ζ zero is identical to the classical plasticity problem. the above equation is nonlinear and must be solved numerically. the above becomes ∆t ΣT R − Yn − 2 [G + 1 (Hiso + Hkin )] λn+1 3 m − ζ λn+1 = 0 (10.101 implies a rate dependency. The stress. with results depending on time durations for applying and changing loads on the body.CHAPTER 10. plastic strain.

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

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)

11. The method may be made computationally viable by making the determination of λA an iterative algorithm.CHAPTER 11.35) and. 11. ill conditioning for the solution to the linear equations will result. In order to yield a solution which provides an adequate satisfaction of the constraint equation.31) where λA is the augmented term. Furthermore. fairly large values for the penalty parameter should be used (generally on the order of about half machine precision. The variation for λA gives the constraint equation g(u) = 0 (11.34) which may be used to compute λ.33) ΛT g(u) dΩ A Ωc The Euler equation for the variation of λ gives the equation g(u) − 1 λ = 0 k (11.30 which are several orders larger than components appearing in the stiﬀness. thus. the constraint may be violated by an unacceptable amount. the augmented Lagrangian method is introduced as an alternative. KIJ . the constraint equation is satisﬁed independently of the value of the penalty parameter. The variation to Eq. The values used then yield stiﬀness modiﬁcations for the second term on the right hand side of Eq. 106 or 107 ). 11.g. Consequently. but with λA used as the multiplier. Using. If the values are too large.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. iterative solutions become very diﬃcult for these large penalty values. AUGMENTED LAGRANGIAN FORMULATIONS 97 The above solution process deﬁnes a perturbed Lagrangian form of the penalty solution process. k. e. these facts we also note that the algorithm merely reduces to the original Lagrange multiplier method. In the Uzawa algorithm we introduce an outer iteration loop for the augmentation. λA ) = Ωc (λ + λA )T g(u) dΩ − Ωc 1 T λ λ dΩ k (11. if too small. and we also conclude that λ must vanish at the solution. The Uzawa algorithm is the simplest algorithm which may be considered. λ. an alternative approach is needed. For each step in the analysis we assume: ..31 gives dΠcη dη = η=0 Ωc ΛT g(u) − + 1 λ k dΩ + Ωc UT GT (λ + λA ) dΩ (11. In the next section.32) (11.

4. A 2. 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.CHAPTER 11. For each time. set j to zero and take the initial value of the augmented multiplier as λA β(j) = λβ (tn ) A (11. If the constraint is satisﬁed to within a speciﬁed tolerance. 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). The number of non-linear iterations will decrease for the later augmented steps since the violation in the constraint is becoming less and less. tn+1 .39) Pλ α In the above the iteration aspects for the incremental problem are not shown. 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. After the iteration for the incremental problem converges update the augmented parameters using β(j+1) β(j) λA = λA + λβ (11. Check convergence for the augmented step. The larger k is made the more rapid the convergence. Let λβ (0) = 0. Use of values with this range in magnitudes leads to 3-6 augmented steps for most problems. If not converged increase the j counter and repeat steps 2 and 3. Let j be the augmentation iteration counter. 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. not quadratic as in a Newton solution. Thus.37) (11. The convergence rate for the augmented iteration is generally linear. .41) is computed using the converged solution from step 2.36) where the dependence on the n+1 step on the left side is implied. 3.40) where λβ = Kλ αβ −1 Pλ α (11.38) (11. AUGMENTED LAGRANGIAN FORMULATIONS 98 1.

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

mixed.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 (12.1) ˜ where σ is computed for a displacement.2) ¨ in which u is the acceleration vector. or enhanced method as described in previous chapters. In order to extend the variational equation to accommodate transient analysis.4) 100 .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. the body force vector. bv .Chapter 12 Transient Analysis of Non-Linear Problems 12. is replaced by ¨ bv ← bv − ρ u (12. 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.

velocities. given the initial conditions.5) (12. ¯ u(0) = d0 .12) . the above equation is a non-linear set of ordinary diﬀerential equations. in the total matrix form as ˜ ¨ R = F − P(σ) − M u (12. the solution at the ﬁrst increment may be determined.2 Newmark Solution of Momentum Equations In this section we illustrate the solution of Eq.CHAPTER 12. The Newmark method is a one-step method which may be used to advance a solution from time tn to time tn+1 . The Newmark method uses approximations to the displacements.6) where MIJ is the mass matrix for the problem. consequently.7) the residual equation becomes ˜ ¨ RI = FI − PI (σ) − MIJ uJ or.11) (12. TRANSIENT ANALYSIS 101 The last term is the inertia contribution to the momentum equation.10) ¯ ¯ where d0 and v0 are the initial displacement and velocity vectors. ˙ ¯ u(0) = v0 (12.9) (12. The method is self starting. and accelerations and these are given as: un ≈ u(tn ) ˙ vn ≈ u(tn ) (12.9 by a time marching process using the classical Newmark method of solution [14].8) In general. If we deﬁne ˜ PI (σ) = e Ωe ˜ BI σ dΩ (12. 12. 12. For continuum problems the acceleration is computed from the isoparametric interpolations as ¨ ¨ u = NJ (x) uJ (t) thus. by ignoring the nodal indices. as described in the next section. The practical solution of the equations is accomplished using a time marching scheme. the inertia term may be written as ¨ ¨ NI ρ u dΩ = MIJ uJ e Ωe (12.

an+1 . however.19) 1 − β) an + β an+1 2 (12. involves only the mass matrix.15) (12. Accordingly.13) The initial state is completed by solving the residual equation at time zero. in general the method is only conditionally stable and very small time steps are needed. ˜ Rn+1 = Fn+1 − P(σ n+1 ) − M an+1 = 0 (12. A β of zero leads to a formulation which is called explicit. whereas 2 for γ > 1 numerical dissipation is introduced. For β non-zero. for β = 0.25. TRANSIENT ANALYSIS and ¨ an ≈ u(tn ) 102 (12. 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.14) In order to advance the solution to the next time it is necessary to recast the problem in an iterative form. For γ = 1 there is no numerical dissipation.CHAPTER 12. the solution for the acceleration. Accordingly. 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 ] (12.17) . where for no damping. the method for linear problems is unconditionally stable. For example. The values of β control primarily 2 the stability but also inﬂuence the form of the matrix problem. 12. The advantage of implicit solutions is improved stability.18) in which β and γ are numerical parameters which control the stability and numerical dissipation. ˜ 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. ¯ v0 = v0 (12.25 should not be used since they are only conditionally stable with allowable time steps not much larger than the explicit scheme. This method is commonly called trapezoidal rule or constant average acceleration. respectively. The advancement of a solution from one step to the next is completed by combining Eq. Values of β less than 0. where quite large time steps may usually be taken.16) (12.18 with the momentum equation written at time tn+1 . This involves selecting appropriate values for the variables to initiate . 12.17 and Eq. For a diagonal mass this solution step is very eﬃcient.

un+1 . as the primary unknowns. With the choice Eq.25) γ (i+1) (i) (i+1) vn+1 = vn+1 + dun+1 (12. the velocities.30.21) 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. and updating of the variables. an+1 .20. For the explicit case the only viable choice is accelerations. TRANSIENT ANALYSIS 103 the step. and M is the mass matrix introduced above. 12. In the sequel we will address the implicit case and use the displacements.29) = − (i) ∂R ∂R ∂v ∂R ∂a + + ∂u ∂v ∂u ∂a ∂u γ 1 Ct + M (12.20) Any other choice may perturb the displacements in such a way to cause false inelastic values. 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. This vector may be the displacements. solution of the linearized equations. Kt = Kt + .28) (12. which impede convergence of the Newton method. Ct is a tangent damping matrix. 12.22) = β ∆t2 dan+1 = γ ∆t dan+1 (i+1) (12.30) β ∆t β ∆t2 In Eq.23) (12.26) β ∆t and 1 (i+1) (i) (i+1) an+1 = an+1 + du (12. Kt is the tangent stiﬀness matrix as computed for the quasi-static problem. 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. vn+1 . or the accelerations. un+1 .CHAPTER 12. For an implicit solution it is best to select the initial value for the iterate as ( un+1 0) = un (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. 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. linearization of the equations.

To reduce the properties to a single parameter.31 gives the tangent matrix K = for use in the Newton method K dun+1 = Rn+α (12.36) .34) α+γ = are employed. 3 2 (12. vn+α ) − M an+1 = 0 where the displacement at the intermeiate point is given by un+α = (1 − α) un + α un+1 and the velocity by vn+α = (1 − α) vn + α vn+1 (12.3 Hilber-Hughes-Taylor (HHT) Algorithm The Newmark algorighm given in the previous section can be altered by considering the residual for the momentum equation to be given by Rn+α = Fn+1 − P(un+α . for short.37) 1 αγ M+ C + αK 2 β∆t β∆t (12.35) Linearization of 12.33) (12.32) (12.31) In the above tn+α = (1 − α)tn + αtn+1 . the relations. α2 4 β= and (12. This algorithm is called the Hilber-HughesTaylor α-method or. TRANSIENT ANALYSIS 104 12. the HHT-method [9] and has been analysed extensively for stability and dissipative properties by Hughes [12].CHAPTER 12.

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

4) is used. FINITE DEFORMATION 106 When common origins and directions for the coordinate frames are used.8) F = ∂X subject to the constraint J = det F > 0 (13.9) to ensure that material volume elements remain positive. Accordingly.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 . 3 A where δaA is a Kronecker delta quantity such that δaA = In component form we then have xa = δaA XA + ua (13. In the above 1 is a rank two shifter tensor between the two coordinate frames. a. may be introduced as the change between the two frames. The determinant of the deformation gradient maps a volume element in the reference conﬁguration into one in the reference conﬁguration. and is given by (13.12) (13.5) 1 = δa Aea ET . x = 1X + u (13.7) 1 if a = A 0 if a = A (13.6) A fundamental measure of deformation is described by the deformation gradient relative to X given by ∂φ (13. A = 1. u.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 conﬁguration. a displacement vector. that is dv = det F dV = J dV (13.CHAPTER 13. 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.

15) B and δAB is a Kronecker delta for the reference conﬁguration. E (do not confuse with the base vectors).18) (13. is introduced as C = FT F (13.21) (13. FINITE DEFORMATION 107 reference or the current conﬁgurations. e.CHAPTER 13. is introduced as 1 E = (C − 10 ) (13.22) b and δab is a Kronecker delta for the current conﬁguration.20) where 1t is the rank two identity tensor with respect to the current conﬁguration and is given by 1t = δab ea eT (13. C. the right Cauchy-Green deformation tensor. Accordingly. 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. . expressed as b = F FT The Almansi strain tensor. for the reference conﬁguration. may be expressed in terms of b as e = 1 1t − b−1 2 (13. b.14) 2 where 10 is the rank two identity tensor with respect to the reference conﬁguration and is given by 10 = δAB EA ET (13.19) (13.16) 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 (13.13) Alternatively the Green strain tensor.17) In the current conﬁguration a common deformation measure is the left Cauchy-Green deformation tensor.

32) .28) (13. S. 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.2 Stress and Traction Measures Stress measures the amount of force per unit of area. ds = J N · C−1 N and (13. They are related through the determinant of the deformation gradient as τ = τab ea eT = J σ = J σab ea eT (13. 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. respectively. and the Kirchhoﬀ stress.24) where n is an unit outward pointing normal to a surface deﬁned in the current conﬁguration.30) where N is an unit outward pointing normal to the reference surface.26) (13. This form of the traction may be related to reference surface quantity through force relations deﬁned as t ds = t0 dS (13. The Cauchy stress.25) (13. remains related to the current conﬁguration forces. 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. thus. and t0 is traction on the reference conﬁguration. In ﬁnite deformation problems care must be taken to describe the conﬁguration to which stress is measured. Note that the direction of the traction component is preserved during the transformation and. P. The reference conﬁguration traction is deduced from the ﬁrst Piola-Kirchhoﬀ stress through t0 = P N (13. σ.27) (13.31) (13.23) b b The second Piola-Kirchhoﬀ stress.CHAPTER 13. are measures deﬁned with respect to the current conﬁguration. τ .29) where ds and dS are surface elements in the current and reference conﬁgurations. the ﬁrst Piola-Kirchhoﬀ stress. FINITE DEFORMATION 108 13.

that is.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.34) ρ v dv (13. for a body force per unit mass. ρ bm dv + Ω ∂Ω td s = Ω ˙ ρ v dv (13. to the rate of ˙ change of the body momentum.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. FINITE DEFORMATION 109 13. both for the current conﬁguration.CHAPTER 13. R. (13. bm the resultant force. The mass density in the current conﬁguration is related to the reference conﬁguration mass density.39) . acting on a body is given by ρ bm dv + Ω ∂Ω t ds = R (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 bm = ρ v This relation is also called the local equilibrium equation for a body. The balance of linear momentum may be expressed by integrating the surface and body loads over the body. Accordingly. through ρ0 = J ρ The total linear momentum of the body is given by p = Ω (13.36) Introducing the relationship between traction and stress and using the divergence principal leads to the balance of linear momentum relation ˙ [div σ + ρ (bm − v)] dv = 0 Ω (13. rho0 . div σ = ∂σab eb ∂xa (13. R. p. Accordingly.33) where rho is the mass density per unit volume and ∂Ω is the surface area of the body.37) where div is the divergence with respect to the current conﬁguration.

41) where the deﬁnitions for rho. thus.46) 13. has only six independent components. 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. FINITE DEFORMATION 110 Similar relations may be constructed for the balance of angular momentum and lead to the requirement σ = σT (13. leads to the corresponding requirement on P F PT = P FT (13. Accordingly. a = 1. The balance of momentum may also be written for the reference conﬁguration using results deduced above. At each point on the boundary one (and only one) boundary condition must be speciﬁed for all three . σ and n ds in terms of reference conﬁguration quantities have been used. we may write the integrals with respect to the reference body as ρ0 bm dV + Ω0 ∂Ω0 t0 dS = Ω0 ˙ ρ0 v dV (13.44) Div P = ∂XA We also note that the symmetry of the Cauchy stress tensor.45) and subsequently to the symmetry of the second Piola-Kirchhoﬀ stress tensor S = ST (13.CHAPTER 13. Using the divergence principle on the traction term leads to the result ˙ [Div P + ρ0 (bm − v)] dV = 0 Ω0 (13.40) that is. 2. ea .4 Boundary Conditions The basic boundary conditions for a solid region consist of two types: displacement boundary conditions and traction boundary conditions. σ. the Cauchy stress tensor is symmetric and.43) In these relations Div is the divergence with respect to the reference conﬁguration coordinates ∂PaA ea (13. The conditions are usually given in terms of their components with respect to a local coordinate system deﬁned by the orthogonal basis. 3.42) which has the local form ˙ Div P + ρ0 bm = ρ0 v (13.

Generally. Thus. components of the position vector. for constitutive equations the initial values for internal variables which evolve in time must be given.50) In general.5 Initial Conditions Initial conditions describe the state of a body at the start of an analysis.CHAPTER 13. x. FINITE DEFORMATION 111 directions of the basis ea .51) 13.48) (13. Accordingly. ∂Ωt . u = x − 1 X = ua ea (13. 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. ∂Ωu .52) (13. In addition.47) for each point on the displacement boundary. 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. boundary conditions may now be given for each displacement component by requiring ua = u a ¯ (13. these conditions may be a mixture of displacement and traction boundary conditions. (13. 2 A speciﬁed quantity is indicated by a superposed bar.49) deﬁne components of the displacement with respect to the prime coordinates. The second type of boundary condition is a traction boundary condition. u. (¯ ·). . For displacement boundary conditions. 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. The boundary condition may also be expressed in terms of components of the displacement vector. the boundary condition is non-linear unless points in the reference conﬁguration can be identiﬁed easily (such as ﬁxed points). Using the orthogonal basis described above.

FINITE DEFORMATION 112 The initial conditions for the kinematic state consist of specifying the position and velocity at time zero. 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 . 0) = d0 (X) and ˙ ¯ v(0) = φ(X. the second Piola-Kirchhoﬀ stress tensor is computed as ∂W (13.CHAPTER 13. For a strain energy density expressed in terms of the right Cauchy-Green deformation tensor. 0) = v0 (X) are speciﬁed at each point in the body.59) 1 2 1 (IC − tr C2 ) = (CKK CLL − CKL CLK ) 2 2 (13. Thus. IIC . As an example. J) (13.61) ∂ IC ∂ C ∂ IIC ∂ C ∂J ∂C .60) (13. ¯ x(0) = φ(X.Finite Elasticity The equations are completed by specifying a material constitution. we consider a ﬁnite deformation form for hyperelasticity. the stress is computed as ∂ W ∂ IC ∂ W ∂ IIC ∂W ∂J S = 2 + 2 + 2 (13. Accordingly.55) (13. The initial conditions for stresses are speciﬁed as ¯ σ(0) = σ 0 at each point in the body. (13. we postulate the existence of a strain energy density function. C. Thus. from which stresses are computed by taking a derivative with respect to a deformation measure.56) S = 2 ∂C For an isotropic material the strain energy density depends only on the three invariants of the deformation.53) 13.57) (13.58) We select J instead of IIIC as the measure of the volume change.6 Material Constitution .54) (13. W .

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 (13.65) The second Piola-Kirchhoﬀ stress may be transformed to the Kirchhoﬀ stress using Eq. 13.66) As an example.63) (13. and gives τ = 2 ∂W ∂W ∂W b + 2 (IC b − 1t ) + J 1t ∂ IC ∂ IIC ∂J (13. Substitution into Eq.65 gives S = 2 µ (10 − C−1 ) + λ J (J − 1) C−1 (13.69) Some formulations require computation of the elastic moduli for the ﬁnite elasticity model.71) The spatial elasticities related to the Cauchy stress. J) = µ (IC − 3 − 2 ln J) + 1 λ (J − 1)2 2 (13. σ.25.67) 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. are obtained by the push forward cijkl = 1 FiI FjJ FkK FlL CIJKL J (13.64) (13. we consider the case of a Neo-Hookean material which includes a compressibility eﬀect.72) . e 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. The elastic moduli with repect to the reference conﬁguration are deduced from C= 4 I ∂2W ∂C ∂C (13.70) (13.62) (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. The strain energy density is expressed as W (IC .

matrix notation is used as much as possible to express the variational equation. Accordingly. As an alternative. it vanishes.75 as3 δΠ = Ω0 ∂W δC dV − ∂C δuT ρ0 bm dV − Ω0 ∂Ω0 t δuT ¯0 dS = 0 t (13. (13. 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 (13.73) Transforming to spatial quantities gives cijkl = λ (2 J −1 ) δij δkl − 2 ( µ − λ (J − 1)) δik δjl J (13. 13.74) Other forms of constitutive equations may be introduced using appropriate expansions of the strain energy density function. however. the notation for a variation to a quantity is written as δ.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. 13. 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. the computations are quite delicate (see [19]). Thus.77) where δu is a variation of the displacement (often called a virtual displacement) which is arbitrary except at the kinematic boundary condition locations. where.75) where ¯0 denotes the speciﬁed tractions in the reference conﬁguration and ∂Ω0 t is the t traction boundary for the reference conﬁguration. in the reference conﬁguration a variational equation is deﬁned from Eq. ueta → u + δu Furthermore. In a ﬁnite element formulation.76) 3 . an elastic formulation may also be expressed in terms of the principal stretches (which are the sqare root of the eigenvalues of C). ∂Ωu .CHAPTER 13. for convenience. the basic element arrays evolve from the balance of linear momentum equations written as a variational equation. Since a virtual displacement is an arbitrary function.

We note that using Eq.78) (13.80) (13.26. Representations with respect to a ﬁxed reference conﬁguration are introduced to simplify the development of the basic relations. Eq. 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 1 δFT P = δFT F S = δC S = δE S (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. . The third and fourth terms of the variational equation represent the eﬀects of body and surface traction loadings.g.82) 2 where symmetry of the second Piola-Kirchhoﬀ stress is noted.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. FINITE DEFORMATION 115 each point in the body as well as the traction boundary conditions.56 and constructing the variation of C. The ﬁrst term side represents the inertial terms. see Hughes [12] or Zienkiewicz and Taylor [26. 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.CHAPTER 13..79) ˙ δuT ρ0 v dV + Ω0 δFT P dV δuT ¯0 dS = 0 t (13. small deformation formulation found in earlier chapters and in ﬁnite element texts (e. 13. by introducing the inertial forces through the body force as ˙ ¨ bm → bm − v = bm − x where v is the velocity vector. For the development considered here it is not important which is selected since ultimately all integrations are carried out over the current conﬁguration. the ﬁrst term reduces to ∂W 1 δC = S δC = δFT P ∂C 2 Furthermore. 27]) except that integrals are performed over the deformed 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.83) (13. the variational equation may be written as δΠ = Ω0 (13. and. 13.

In the above.89) + + Ω tr ( (δu) ∆σ) dv where 1 F ∆S FT (13.86) Note that in the reference conﬁguration the domain. plates and shells and. but other forms also hold) tr (A B) = AIJ BJI (13. thus. The linearization should be considered in a reference conﬁguration representation. Accordingly. which is not true for problems in beams. Ω0 is ﬁxed (i. the second term depends on the material constitution and leads to the material part of the stiﬀness. The linearization may be transformed to the current conﬁguration and expressed in terms of quantities associated with the Cauchy stress.87) We also note that for a continuum problem ∆ (δF) vanishes. Consequently. To solve a boundary value problem the nonlinear equations may be linearized and solved as a sequence of linear problems.CHAPTER 13. whereas.e. additional terms are necessary. Accordingly.90) J The ﬁrst term on the right hand side leads to the geometric stiﬀness term in a ﬁnite element formulation. the trace operation denotes the following step (reference conﬁguration tensors are used as an example. a formulation based upon the second Piola-Kirchhoﬀ stress and written in tensor form is considered for the linearization step. In this section it is expedient to again use a tensor form of the equations instead of the matrix form used above.88) tr (δF S FT dV ) = Ω tr ( (δu) σ (∆u)T ) dv (13.85) which are equivalent forms. as shown above. does not change) which is not true for a formulation considered directly in the current conﬁguration. 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. using δF = and a similar expression for ∆F gives ∆( Ω0 (δu) F (13. ∆σ = .8 Linearized Equations The stress divergence term may be written in many forms. FINITE DEFORMATION 116 13. tr (δFT F S) dV = Ω0 Ω0 tr (δF S FT ) dV (13.

The moduli c are related through J I = F F C F T FT c I (13. Accordingly.9 Element Technology A ﬁnite element discretization may be constructed by dividing the body into ﬁnite elements. 13. and Ωh is the domain covered by all the elements.CHAPTER 13. When used with the deﬁnition of ∆σ this may be transformed to the current conﬁguration as ∆σ = I ∆ c (13. This will be discussed later for a particular constitutive equation. however. It is expressed as 1 ∆ = [ ∆u + ( ∆u)T ] (13.91) where Care the material moduli for the material constitution expressed in the reference I conﬁguration. We note that in general Ωh is an approximation to the domain of the real body since the elements only have mapped polygonal shapes. we have Ω ≈ Ωh = e Ωe (13. e.95) which we note is also identical to the form of the linear problem.97) Using this approximation the variational equation become ˙ δuT ρ v dv + e Ωe Ωe (δu)T σ dv = .96) where Ωe is the domain of an individual element. With this approximation the integrals in the variational equation may be approximated as ( · ) dv ≈ Ω Ωh ( · ) dv = e Ωe ( · ) dv (13.93) In the above ∆ is the symmetric part of the gradient of the incremental displacement.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. in general we seek an expression of the form ∆S = C∆E I (13. FINITE DEFORMATION 117 The material part involves ∆S which is computed for each particular constitutive relation.92) where I are the material moduli expressed in the current conﬁguration.

2.103) (13. respectively.105) where summation convention is implied for the a and b indices. NI (ξ) are shape functions for node I which maintain suitable continuity between contiguous elements and XI are the coordinates for node I. 27]) we may write for a typical element X = NI (ξ) XI . Similarly. I = 1. FINITE DEFORMATION δuT ρ bm dv + e Ωe ∂Ωet 118 δuT ¯ ds t (13.104) (13. nen (13.g.106) Ωe For procedures to construct a lumped mass see either [26.101) Time dependence is included in the nodal parameters for the current position and displacements..99) An approximate variational solution may be developed by writing trial solutions and test functions for the motions and virtual displacements. see [12. Accordingly.10 Consistent and Lumped Mass Matrices Using the above approximations we may discretize the terms in the variational equation for each element. 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 (13.100) where nen is the number of nodes deﬁning an element. 13. . Appendix 8] or [11]. the ﬁrst term becomes ˙ δuT ρ v dv = (δuT )I Ωe Ωe ¨ NI ρ NJ dv 1t xJ (13.CHAPTER 13. Adopting an isoparametric formulation (e. . 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 (13.102) (13. I are node labels for the element.

2 δu3 NI.1 NI.1 = δ (13. The BI matrix describes the transformation from the virtual displacements.1 0 0 0 NI.CHAPTER 13.3 1 δuI = = BI δuI 2 NI.111) Expressing the δ in terms of the virtual displacements gives δ = ∂δu1 ∂x1 ∂δu2 ∂x2 ∂δu3 ∂x3 ∂δu1 ∂x2 + ∂δu2 ∂x1 ∂δu2 ∂x3 + ∂δu3 ∂x2 ∂δu3 ∂x1 + ∂δu1 ∂x3 T (13. FINITE DEFORMATION 119 13. The stress divergence term may now be written as δ T σ dv = (δuI )T BT σ dv (13. .109) T 31 δ 33 2δ 12 2δ 23 2δ (13.108) σ11 σ22 σ33 σ12 σ23 σ31 δ 22 T (13.114) ∂x1 has been used for conciseness. the notation ∂NI (13.115) 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. 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 (13.3 NI.2 NI.110) the stress divergence term may be written as δ Ωe T σ dv (13.3 0 NI.113) In the above.112) Using the interpolations for the virtual displacements in each element leads to the matrix relation NI.2 0 I δu 0 0 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 (13.1 0 I 0 NI.

FINITE DEFORMATION 120 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.standard B matrix formulation The material tangent matrix is deduced from the term tr ( (δu) ∆σ) dv = Ωe Ωe tr (δ I ∆ ) dv c (13.122) .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.119) which is evaluated for a typical element. Furthermore. substitution of the ﬁnite element interpolations into the incremental strain term leads to the result in matrix form ∆ = BJ ∆uJ (13.13 Material tangent matrix .120) where D denotes the material moduli in the current conﬁguration given in the matrix representation introduced for the linear problem. tr ( (δu) σ Ωe (∆u)T ) dv = (δuI )T Ωe T tr ( NI σ NJ ) dv ∆uJ (13.121) Thus. Accordingly. the material tangent is computed from δ Ωe T D ∆ dv = (δuI )T Ωe BT D BJ dv ∆uJ I (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. In matrix notation the right hand side becomes tr (δ I ∆ ) dv = c Ωe Ωe δ T D ∆ dv (13.118) 13.CHAPTER 13.

the variational equation may be written as (δuI )T e ¨ MIJ xJ + Ωe BT σ dv − fI I = 0 (13.126) where fI is the sum of the body and surface traction terms. the variational equation leads to the discrete balance of linear momentum equations ¨ MIJ xJ + e Ωe BT σ dv − fI I = 0 (13.15 Basic ﬁnite element formulation Accumulating all terms together.127) Since δuI is arbitrary. 13.123) which again is identical to the linear problem except that all steps are performed for the current conﬁguration.124) and the boundary loading is δuT ¯ tds = (δuI )T ∂Ωe ∂Ωet NI ¯ ds t (13.14 Loading terms The right hand side terms may be discretized by introducing the interpolations for the virtual displacement.CHAPTER 13.128) which may be written in the compact matrix form ¨ M x + N(σ) = f (13.129) .125) 13. fI = Ωe NI ρ bm dv + ∂Ωe t NI ¯ ds t (13. Accordingly. the body force may be given as δuT ρ bm dv = (δuI )T Ωe Ωe NI ρ bm dv (13. FINITE DEFORMATION and the material tangent matrix is given by (Km )IJ = Ωe 121 BT D BJ dv I (13.

131) 13. (as described in [19]). a modiﬁed deformation gradient. 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.137) (13. the modiﬁed deformation gradient is based upon a separation into volumetric and deviatoric parts as F = Fvol Fdev (13.16 Mixed formulation In the mixed formulation used. The mixed formulation is used to permit solution of incompressible and nearly incompressible materials.132) The above description is for a standard displacement type formulation.CHAPTER 13. 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. Accordingly. We refer to the method as the standard B-matrix formulation. in a Newton Method we write the momentum equation as ¨ R = f − M x − N(σ) = 0 (13. compressible solutions which can be treated by a standard B matrix formulations. as well as. Thus.130) A linearization of this set of equations gives the result M ∆¨ + Kt ∆u = R u where Kt = Km + Kg (13.135) (13.133) where Fvol measures volumetric part and Fdev the deviatoric part of deformation. yields the solution to a problem. (13. is used.136) (13.138) .134) (13. 122 Solution of this set of equations together with satisfying the material constitution and the displacement boundary conditions.

142) Ω0 Ω0 − Ω0 uT ρ0 bm dV − ∂Ω0t t uT ¯0 dS (13. The modiﬁed deformation gradient may then be constructed by replacing the volumetric part by a mixed treatment. ˜ Π(u.147) . θ.141) 3θ 3 A three ﬁeld variational statement of the problem is completed by adding to the motion. σ. The virtual modiﬁed deformation gradient is now given by δθ 1 ˜ ˜ δF = 1t + ( δu − div δu 1t ) F (13. θ. and δp (1 − Ω θ ) dv = 0 J (13.146) for the relation between the mixed and the determinant of the deformation gradient. we deﬁne ˜ F = θ J 1 3 F (13. θ) dV + p (J − θ) dV (13. p.143) A variational equation including the eﬀects of inertia may be constructed following steps above and written as (see. the mixed pressure.145) for the relationship between the mixed pressure and the trace of the stress. and the modiﬁed Kirchhoﬀ stress.140) ˜ ˜ with similar deﬁnitions for E and b. FINITE DEFORMATION 123 for the deviatoric part. δθ( Ω ˜ trσ p − ) dv = 0 3θ J (13. p) = W C(u. ˜ ˜ In the above.139) as the modiﬁed tensor. [19]) ˙ δuT ρ v dv + Ω Ω ˜ δuT (σ dev + p 1t ) dv = Ω δuT ρ bm dv + ∂Ω δuT ¯ ds (13. The modiﬁed right Cauchy-Green deformation tensor is then computed from ˜ ˜ ˜ C = FT F (13. Accordingly. θ is a mixed representation for the determinant of the deformation gradient.CHAPTER 13. τ .144) t for the linear momentum equation. In the above expression. and mixed determinant of the modiﬁed deformation gradient. φ. are related to the modiﬁed second Piola-Kirchhoﬀ stress by ˜˜˜ ˜ ˜ J σ = τ = F S FT (13. the modiﬁed Cauchy stress.

using the third variational equation.150) BT (σ dev + p 1t ) dv I ˜ (13.148) where I is a rank four identity tensor. A tangent matrix may be computed for the mixed formulation. This may be combined with the deviatoric part of σ to deﬁne the mixed stress. FINITE DEFORMATION 124 ˜ ˜ where S is computed using C as the deformation measure.153) Ωe0 Ωe 3 θ ˜ in each element. p. δθ. The θ may now be ˜ used to deﬁne the modiﬁed deformation quantity and the modiﬁed stress state. x. Finally. and δu. 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. appearing in the above relations may be obtained by ﬁrst computing the mixed volume. where Ωe is the volume of the element in the current conﬁguration and Ωe0 is the volume in the undeformed reference conﬁguration. σ. 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. The spherical part of the stress is given by the ˜ ˜ mixed pressure. p. Thus. not trσ. Details for the construction are included in [19]. The mixed pressure p is computed from trσ using the variational equation given above. The deviatoric part of the ˜ stress.149) A ﬁnite element implementation for the above may be deduced using the isoparametric interpolations given above for X.CHAPTER 13. .152) Ωe0 for each element. In addition interpolations for θ. In the low order elements the above functions are all taken as constant in each element.151) The pressure. the stress in this approach is computed using ˜ σ = p 1t + σ dev (13. in each element. σ may be determined in each element. use of the second variational equation yields the mixed pressure as ˜ 1 trσ p = dv (13. p. σ dev . and δp must be given. Accordingly. u.

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

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

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

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

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

APPENDIX A. pcomp.’Axisym’/ 130 logical character real*8 data c Input material parameters d(4) = 1.d0 elseif(pcomp(name.d(2).d(2).e12.5/5x.5/ 5x.true.’Linear Heat Conduction Element’// 5x. rflag name*15.’axis’.’cond’.td.4)) then d(4) = 1.1) if(pcomp(name.lt.1.wlab(int(d(4))) write(iow.d0 elseif(pcomp(name.td(1) wlab/’ Plane’.wlab(int(d(4))) d(2) = d(2)*d(3) d(5) = 2 ! number of quadrature points/direction 2000 & & format(5x.2: Input Routine for Heat Transfer Element .’ Analysis’) end Table A.2000) d(1).4)) then d(4) = 2.e12. HEAT TRANSFER ELEMENT subroutine inpt02(d) implicit include none ’iofile. do while(rflag) errck = tinput(name.5/5x.d(3).’dens’. endif end do ! while if(ior.4)) then d(3) = td(1) elseif(pcomp(name. wlab(2)*6 d(*).’plan’.2000) d(1).’Density ’.d0 rflag = .’ ’.’spec’.’Conductivity ’.4)) then rflag = .’Specific Heat’.h’ errck.d(3).a6.false.4)) then d(2) = td(1) elseif(pcomp(name.0) write(*.4)) then d(1) = td(1) elseif(pcomp(name. tinput.e12.

lint.xsj.lint call shape(sg(1.APPENDIX A.tdot) if(nint(d(4)).nst.r(ndf. a1.nel.nel-1 s(i*ndf+1.j+1) .a3.j*ndf+1) = s(i*ndf+1.i+1))*ctan(1) & + a3*shp(3.9).a2*gradt(2) .nel) do j = 0.a2.l).i+1)*ctan(2) end do end do end do end Table A.l) a3 = d(2)*shp(3.a1*gradt(1) . tdot.9). and residual l = nint(d(5)) call int2d(l.*) shp(3.ul.ndf.l).lint.ndm.eq.xl.a3*tdot do i = 0.3: Stiﬀness for Heat Transfer Element .j*ndf+1) & + (a1*shp(1.sg(3.j+1) = r(1. gradt(2).*).xl(ndm.h’ ndf. ix(*) xsj.) call flux02(d.j+1)*xsj*sg(3.shp.nel.s.xl.ndm.xl.j+1)*xsj*sg(3.ix.flux(2) Compute tangent matrix (linear).false.ul(ndf.ix.r. gradt.nel-1 a1 = d(1)*shp(1. HEAT TRANSFER ELEMENT 131 subroutine stif02(d.ndm.ul..h’ ’eltran.l) r(1.shp.sg(2.i+1) + a2*shp(2.flux. l. i.nst) implicit include include integer real*8 real*8 real*8 c none ’eldata.l) a2 = d(1)*shp(2.ndm.ndf.*).2) xsj = xsj*radi02(shp. radi02 d(*).s(nst.j.sg) do l = 1.*).j+1)*xsj*sg(3.

ul.yy.2001) o.flux. radi02 d(*).ndf. gradt.xl(2.0) then write(iow.lt.shp.0 .xx.’element flux’//’ elmt matl 1-coord .3.ix.*).head mct = 50 endif xx = radi02(shp.2001) o.h’ ’cdata.xl.yy.nel.ma.1 if(mct.ndf.tdot) mct = mct . ix(*) xx.1).ul(ndf.h’ ’eldata.gradt if(ior.le.2002) n.flux.0 .) call flux02(d.0d0. xsj.xl.0.flux(2).lt.ndm.shp(3. tdot.3) 2-coord’ 2-grad’) end Table A.ndm.and.*).h’ ’fdata.xsj.false.ndm.ma.1).ndm.head if(ior.9) Compute thermal gradient and heat flux call shape(0.nel) write(iow.xx.gradt(2).20a4//5x.nel.h’ ’iofile.ul.and.yy. HEAT TRANSFER ELEMENT 132 subroutine strs02(d.4: Output Routine for Heat Transfer Element .ndm) implicit include include include include include integer real*8 real*8 c none ’bdata.4e12.gradt 2001 & 2002 format(a1.2f9.xl(1.h’ ndf.0d0.’ 1-flux 2-flux 1-grad format(2i5.xl(ndm.2002) n.flux.nel) yy = radi02(shp.ix.APPENDIX A.shp. pfr) write(*.. pfr) write(*.

xl.9) Lumped projection routine l = max(2.sg(2.tdot) xsj = xsj*sg(3.5: Flux Projection Routine for Heat Transfer Element .numnp.shp.false.flux. HEAT TRANSFER ELEMENT 133 subroutine stcn02(ix.d(*) gradt(2).lint.shp.2) = st(ll.l) do j = 1.) call flux02(d.nint(d(5))) call int2d(l.dt.ndf.lint. ix(*) xsj. gradt.ndm.shp(3.1) + flux(1)*xg st(ll.xl.shp.ul.nel.0) then xg = xsj*shp(3.*).sg(3.1) = st(ll.d.j) dt(ll) = dt(ll) + xg st(ll.l).numnp) implicit none integer real*8 real*8 real*8 c ndf.ul.st.APPENDIX A.l.st(numnp..4).sg) do l = 1. j.*).ul(ndf.nel.xl(ndm.gt.l).xsj. tdot dt(numnp).nel.2) + flux(2)*xg endif end do end do end Table A.ndf.xg.flux(2).nel ll = iabs(ix(j)) if(ll.ndm.lint call shape(sg(1.ndm.ll.nel.ix.*).

ul. gradt.tdot) implicit include integer real*8 real*8 none ’cdata.i)*xl(1.0d0 tdot = 0.xl.nel radi02.APPENDIX A.nel.0d0 do i = 1.gradt(*).ndf. HEAT TRANSFER ELEMENT subroutine flux02(d.1) tdot = tdot + shp(3.0d0 gradt(2) = 0.flux(*) 134 gradt(1) = 0.nel.shp.i)*ul(1.*).ul(ndf.ndm.nel radi02 = radi02 + shp(3.i) end do end Table A.nen.i. xl(ndm.h’ ndm.i)*ul(1.flux.1) gradt(2) = gradt(2) + shp(2.ndm.*) Compute element coordinate value radi02 = 0.i)*ul(1.7: Coordinate in Element . shp(3.*).i.6: Thermal Gradient and Flux function radi02(shp.d0 do i = 1.*).nel) implicit none integer real*8 c i.i. i tdot d(*).nel gradt(1) = gradt(1) + shp(1.4) end do flux(1) = -d(1)*gradt(1) flux(2) = -d(1)*gradt(2) end Table A.shp(3.

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

Appendix B Solid Elements B. The strain-displacement matrices for each node are given by: 1.4) 136 .x 0 0 BI = NI. Two dimensional plane problems = x y γxy γyz γzx 0 0 NI.2) (B.z NI.1 Displacement elements Displacement elements are computed using the virtual work equation written in terms of assumed element displacments.z NI. All elements for continuum (solids) analysis use isoparametric displacement ﬁelds expressed as u= I NI (ξ) uI (B. Three dimensional problems = x y z NI.x T (B.y 0 0 NI.z 2. Computation of the derivatives appearing in the strain-displacement matrices is performed as described in Appendix D.3) z γxy T (B.1) where NI (ξ) are shape functions and uI are nodal displacements.y NI.x 0 NI.y 0 NI.

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

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

**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.

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

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

43) D. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS Next by multiplying (again from the right) by h. 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. the derivatives with respect to the global coordinates. u.44) where the biI are constant over the entire element and are computed by the conventional expressions at the center of the element.46) ξh = ξ1 Furthermore. x.41) Ih = h = i=1 x h bi + 4 Γ i where2 xh = xT h i i Thus. and when the coordinate. the factor uh . the gradient with respect to the natural coordinates is given by ξ2 (D.42) x h bi ] i i=1 (D.47) The factor xh is sometimes called an hour glass shape.40) (D. i .3 Derivatives of Alternative Formulation Using the alternative expression for the shape functions. 2 (D. xi . we get hT h = 4 2 144 (D. deﬁnes the magnitude of the hour glass mode.45) For the speciﬁc functional expression for h. is replaced i by the displacement.APPENDIX D. the parameters for Γ are computed as 1 Γ = [h − 4 It remains to compute the bi . The derivatives of the function h may also be computed using the chain rule and are given by x h = ξ h J−1 (D. are given by ∂h ∂NI = biI + ΓI ∂xi ∂xi (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.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. Recall that the derivative of a global coordinate with respect to a natural coordinate has a constant and a linear part.48) where j0 is the value of the jacobian determinant evaluated at the element center. J.51) (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. ISOPARAMETRIC SHAPE FUNCTIONS FOR ELEMENTS 145 where j(ξ) is the determinant of the jacobian transformation matrix. The jacobian determinant at the center of the element is computed to be j0 = (J0 )11 (J0 )22 − (J0 )21 (J0 )12 (D.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.53) (D.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.

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

APPENDIX E. PROPERTIES FOR J2 PLASTICITY MODELS

147

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

E.1

Example 1

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

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

(E.13)

E.2

Example 2

H2 = C I + D (I − n nT ) (E.15)

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

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

148

(E.18)

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

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

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

(E.21)

(E.22)

(E.23)

(E.24)

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

**Appendix F Matrix Form for Equations of Solids
**

F.1 Stress and Strain

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

T

(F.2)

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

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

11) F. MATRIX FORM FOR EQUATIONS OF SOLIDS The 6-component form may be related jector matrix.5) (F.APPENDIX F.9) This permits the independent components of strain to be written in a 6-component matrix form as T = (F. P.10) 11 22 33 γ12 γ23 γ31 where γiij are the engineering components of the shearing strain given by γij = 2 ij (F.7) and reordered into the 9-component vector as = T 11 22 33 12 21 23 32 31 13 (F.8) Strain-displacement relations give symmetry of strain as ij = ji (F.12) σ = s + mp .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. 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.6) = (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. 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. MATRIX FORM FOR EQUATIONS OF SOLIDS and 151 1 m εv 3 where p and εv are the pressure and volume change.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.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.17) (F. respectively. for the stress we have the two relations 1 σ = s + m mT σ (F.14) (F.25) 3 3 . Accordingly.APPENDIX F.21) The 6-component projector m is likewise related to its 9-component counterpart through m = PT m = 1 1 1 0 0 0 T (F.19) 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.15) (F.23) 3 and 1 (F.13) (F.18) (F.

APPENDIX F. (F.29) F. respectively.30) where Cijkl are the elastic moduli and possess the minor symmetries Cijkl = Cjikl = Cijlk (F.26) where I and I are identity matrices of size 9 and 6.35) where D is a 9 × 9 matrix of elastic constants and D is a 6 × 6 matrix of elastic constants.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.31) From notions of hyperelasticity where stress is deduced from the stored energy function W ( ) as ∂W σij = (F.33) .27) Similarly for strains we have the deviatoric relations e= − and e= − 1 m mT = Idev 3 1 m mT = Idev 3 (F.28) (F. MATRIX FORM FOR EQUATIONS OF SOLIDS and s=σ− 152 1 1 m mT σ = I − m mT 3 3 σ (F. We deﬁne the two deviatoric projectors as Idev = I − 1 m mT 3 and Idev = I − 1 m mT 3 (F.34) (F.3 Linear Elastic Constitutive Equations Let us now consider the relations for linear elastic constitutive equations. In index notation these are expressed as σij = Cijkl kl (F.

(F.1.APPENDIX F.37) F.40) where Iijkl is the rank-4 tensor identity.39) we obtain the tensor form of the elastic moduli as Cijkl = λ δij δkl + 2 µ Iijkl (F. MATRIX FORM FOR EQUATIONS OF SOLIDS 153 Construction of D follows directly from Cijkl using the index maps shown in Table F.3.36) 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.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.2: Matrix and tensor index maps (F.42) .38) where in index form εv = kk .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. Writing the relationship for the constitution as σij = Cijkl kl (F.2. This may be directly related to a matrix form as (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. 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.

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

Sign up to vote on this title

UsefulNot useful- FEAP Theory
- IFEM.Ch11
- Lecture 2
- Introduction to Finite Element Method
- Wu Xiaohe 200405 PhD
- Lagrange Equation
- Lecture 5 Variational Principle
- Note for Mac
- Effect of Temperature Variation and Shrinkage on Circular Tanks
- IMSC002 Elasticity
- Shake User Manual
- Gate Engineering Syllabus
- 14_05-03-0148
- Chap-01
- Abq Plasticity Tutorial Ver 611
- MeasurementForControl V3_01
- 1D Elasticity 02 Elastodynamics
- Education
- sct
- Production engineering syllabus
- Analysis of PDC Wobbling & DS Buckling
- Straingauge 141220080949 Conversion Gate01
- Monge Kantorovich.survey (1)
- 9709_w04_ms_3
- CHPR4406 LectureS
- BE - B.tech Syllabus
- SPE-91101-MS
- 252Ch.2WORDPROBLEMS
- sol1
- <HTML><HEAD> <TITLE>Refresh</TITLE>ecp 4 </HEAD> <BODY> <FONT face="Helvetica"> <big><strong></strong></big><BR> </FONT> <blockquote> <TABLE border=0 cellPadding=1 width="80%"> <TR><TD> <FONT face="Helvetica"> <big>Refresh (dynamic_bypass_reload)</big> <BR> <BR> </FONT> </TD></TR> <TR><TD> <FONT face="Helvetica"> Click <a href="">here</a> if you are not automatically redirected. </FONT> </TD></TR> <TR><TD> <FONT face="Helvetica"> </FONT> </TD></TR> <TR><TD> <FONT face="Helvetica" SIZE=2> <BR> For assistance, contact your network support team. </FONT> </TD></TR> </TABLE> </blockquote> </FONT> </BODY></HTML>
- Method of Finite Elements