l'li'St r · • JY

•
Prentice Hall International (UK) Ltd
Campus 400, Maylands Avenue, Hemel Hempstead
Hertfordshire HP2 4RG
I Contents I
=-
- A division of
Simon & Schuster International Group

©Prentice Hall International (UK) Ltd, 1992

All rights reserved. No part of this publication rna~ be

reproduced, stored in a retrieval system, or transr~utted,
in any form, or by any means, elec~ronic: mechan~cal,
photocopying, recordmg or otherwts~, wtthout pnor
permission, in writing, from th_e pubhsher. .
For permission within the Umted States ?f Amenca
contact Prentice Hall Inc., Englewood Cltffs, NJ 07632. Preface xi
Typeset in 10/ 12 pt Times Suggestions to the instructor xiii
by p & R Typesetters Ltd, Salisbury, Wilts
Acknowledgements XV
Printed and bound in Great Britain
by Dotesios Ltd, Trowbridge, Wilts 1 Introduction 1
1.1 Basic description 1
Library of Congress Cataloging-in-Publication Data
1.2 Some specific applications 7

2 Matrix algebra 11
Ottosen, Niels Saabye. . 2.1 Definitions
Introduction to the finite element method/Ntels Saabye Ottosen 11
and Hans Petersson. 2.2 Addition and subtraction 13
p. em. . 2.3 Multiplication 13
Includes bibliographical references and mdex. 2.4 Determinant 16
ISBN 0-13-473877-2 II. Title. 2.5 Inverse matrix 18
l. Finite element method. I. Petersson, Hans.
2.6 Linear equations: number of equations is equal to number
TA347.F5088 1992 91-41612
620'.001'5153S- dc20 of unknowns 19
CIP 2.7 Linear equations: number of equations is different from
number of unknowns 23
2.8 Quadratic forms and positive definiteness 23
2.9 Partitioning 24
British Library Cataloguing in Publication Data
2.10 Differentiation and integration 25

Ottosen, Niels Saabye 3 Direct approach - truss analysis 27

Introduction to the finite element method.
I. Title II. Petersson, Hans 3.1 Simple structure consisting of springs 27
515 3.2 Truss analysis- local and global coordinates 37
ISBN 0-13-473877-2
4 Strong and weak formulations - one-dimensional heat flow 48
1 2 3 4 5 96 95 94 93 92 4.1 One-dimensional heat equation- strong form 48
4.2 Axially loaded elastic bar 52
4.3 Flexible string 54

v
vi Contents Contents vii
Weak form of one-dimensional heat flow 56 9 FE formulation of one-dimensional heat flow
4.4 157
4.4.1 Advantages of the weak formulation compared with the 9.1 Global FE formulation 157
strong form 59 9.1.1 Example 1 164
62
4.4.2 Alternative derivation of heat equation 9.1.2 Further properties of the stiffness matrix 170
64 9.1.3 Systematic consideration of boundary conditions
4.5 Concluding remarks 171
9.1.4 Example 2 174
9.1.5 Evaluation of the force vector - fulfilment of the global
5 Gradient, Gauss' divergence theorem and the Green-Gauss balance principle
theorem 65 175
65 9.1.6 Evaluation of the load vector- point sources 177
5.1 Gradient 9.2
70 Expanded FE formulation of one element - assembly process 179
5.2 Gauss' divergence theorem and the Green- Gauss theorem 9.2.1 Example 3 183
9.3 FE formulation of one element 184
6 Strong and weak forms- two- and three-dimensional heat flow
76 9.3.1 Example 4 188
Heat flux vector - constitutive relation 76 9.3.2 Use of local coordinate systems
6.1 191
Heat equation for two and three dimensions - strong form 81 9.4 Axially loaded elastic bar
6.2 193
Weak form of heat flow in two and three dimensions 85 9.4.1 Example 5
6.3 195
Other analogous physical problems 87 9.5 Basic features of an FE computer program
6.4 197
9.6 Heat flow with convection 198
9.7 C 0 -continuity 202
1 Choice of approximating functions for the FE method -
90 9.8 Concluding remarks 203
scalar problems
General requirements 91
7.1 10 FE formulation of two- and three-dimensional heat flow
7.1.1 Convergence criteria 91 206
94 10.1 Two-dimensional heat flow 206
7.1.2 Rate of convergence - Pascal's triangle 10.1.1 Example
One-dimensional elements 98 214
7.2 10.1.2 Two-dimensional heat flow with convection
7.2.1 Simplest possible one-dimensional element 98 219
106 10.2 Three-dimensional heat flow 220
7.2.2 Quadratic one-dimensional element
7.2.3 Cubic and quartic one-dimensional elements - Lagrange 11
115 Guidelines for element meshes and global nodal numbering 223
interpolation
118 11.1 FE mesh 223
7.3 Two-dimensional elements
118 11.2 Methods of solution 225
7.3.1 Simple triangular element 11.2.1 Bandwidth
7.3.2 Four-node rectangular element - the Lagrange element 126 225
11.2.2 Systematic Gauss elimination 226
7.3.3 More complicated rectangular and triangular elements - 11.2.3 Global nodal point numbering
serendipity elements 131 230
136 11.3 Estimation of bandwidth 231
7.4 Three-dimensional elements
Summary 138
7.5 12 Stresses and strains 235
12.1 Stresses 235
Choice of weight function - weighted residual methods 142 12.1.1 Plane stress
8 241
Point collocation method 147 12.2 Strains
8.1 243
Subdomain collocation method 149 12.2.1 Plane strain
8.2 247
Least-squares method 151
8.3
The Galerkin method 152 13 Linear elasticity
8.4 248
Example 153 13.1 Symmetry properties
8.5 251
Concluding remarks 156 13.2 Initial strains
8.6 253
viii Contents
Contents ix
13.3 Plane stress 254 17.3.5 Beam element with no distributed loading 330
13.4 Plane strain 255 17.3.6 Evaluation of element load vector - uniform load 331
13.5 Summary of equations for solid mechanics 256 17.3.7 Evaluation of element load vector - concentracted
13.5.1 Comparison of the equations of heat flow and solid force and moment
mechanics 259 332

14 FE formulation of torsion of non-circular shafts 261 18 FE formulation of plates

18.1
335
14.1 Saint-Venant torsion theory 261 Plate theory
335
14.1.1 Differential equation for Saint-Venant torsion theory 261 18.1.1 Equilibrium conditions
335
14.1.2 Boundary conditions 265 18.1.2 Kinematic relations
339
14.1.3 Determination of torque 267 18.1.3 Constitutive relation
340
14.1.4 Soap film analogy 269 18.1.4 Further derivations
341
14.1.5 Discussion of Saint-Venant torsion theory 270 18.1.5 Differential equations fo r plate theory 342
14.2 FE formulation 271 18.1.6 M.omen.ts and shear forces acting on an arbitrary plane 344
14.2.1 Example 276 18.1.7 Discussion of proper boundary conditions 347
18.2 Weak form of equilibrium equation - proper static boundary
15 Approximating functions for the FE method - vector conditions
349
problems 282 18.3 Concentrated shear forces at corners
352
15.1 Convergence criteria 282 18.4 FE formulation
356
15.2 General approximation techniques 284 18.4.1 Completeness and compatibility requirements
360
15.2.1 Three-node element - constant strain element 288 18.4.2 Two simple non-conforming plate elements
361
15.2.2 Four-node rectangle - Melosh element 289
19 Isopararnetric finite elements
16 FE formulation of three- and two-dimensional elasticity 292 364
19.1 Restrictions on the mapping
16.1 Weak form of equilibrium equations - three-dimensional case 292 364
19.2 Four-node isoparametric quadrilateral element
16.2 FE formulation of three-dimensional elasticity 295 366
19.3 Eight-node isoparametric quadrilateral element
16.3 Weak form of equilibrium equations - two-dimensional case 299 370
19.4 Three-dimensional isoparametric elements
16.4 FE formulation of two-dimensional elasticity 302 373
19.5 General isoparametric formulation - convergence requirements
16.4.1 Example 1 306 374
19.6 Integral transformations
16.4.2 Example 2 308 376
19.6.1 Transformation of integrals over one-, two- and three-
17 FE formulation of beams 311 dimensional regions 376
17.1 Beam theory 312 19.6.2 Transformation of boundary integrals 380
19.7 Evaluation of FE equations using isoparametric elements
17.1.1 Equilibrium conditions 312 382
17.1.2 Kinematic relations 313 19.7.1 Example
385
19.8 Need for numerical integrations
17.1.3 Constitutive relation 315 386
17.1.4 Choice of position of x-axis 315
17.1.5 Differential equations for Bernoulli 's beam theory 317 20 Numerical integration
17.2 Weak form of equilibrium equation 318 20.1
387
Newton-Cotes quadrature
17.3 FE formulation 319 389
20.1.1 Example 1
17.3.1 Completeness and compatibility requirements 323 390
20.1.2 Example 2
17.3.2 Simplest possible beam element 324 391
20.2 G auss integration
17.3.3 Evaluation of element stiffness matrix 328 391
20.2.1 Example 3
17.3.4 Evaluation of element boundary vector 329 392
20.2.2 Example 4
394
x Contents
20.3 Introductory remarks on the order of Gauss integration of

I I
394
isoparametric elements
20.4 Reduced integration and spurious zero-energy m?des . 396 Preface
20.4.1 Suggested order of Gauss integration of 1soparametnc
402
elements
403
References
407
Index

This textbook is intended to be used as an introductory text for students in various

fields of engineering. It presents a joint effort by the Divisions of Structural Mechanics
and Solid Mechanics at the University of Lund to combine their experiences in the
teaching of the finite element method.
The text deals with the formulation of the finite element method and the intention
is that the reader should be able to formulate the-finite element approach for arbitrary
differential equations. To facilitate this approach the weak formulation of the
differential equations in question is used in combination with the Galerkin method.
To achieve this objective, and in order to be able to evaluate the physical
importance o the different terms appearing in t he equations, a profound
understanding of the physical significance of the various differential equations
discussed is essential. Therefore, emphasis is given to a detailed derivation of the
differential equations encountered. In this manner, the book serves partly as an
introduction to the various differential equations typically occurring within
engineering mechanics and partly as an introduction to the finite element formulation
of these equations. By this two-fold approach it is hoped that the reader will be left
with a firm grasp of many of those problems that he or she may encounter in his or
her later professional career.
The background required by the student is modest, as only an introductory
knowledge of mechanics, matrix algebra, and differential and integral calculus is
required. All differential equations of the physical problems dealt with are derived,
and the exposition is self-contained with detailed derivations so that the book may
also be used for individual studies.
Most textbooks within this field are concerned either with the derivation of the
equations of mechanics or the formulation of the finite element method itself. We
believe that the present text represents a modern trend in teaching engineering
disciplines, since all the differential equations dealt with are derived before the
corresponding finite element formulation is established. While this approach may be
inconvenient for students of mathematics, it is likely to be very helpful for engineering
stt.dents. In particular, a clear perception of the physical assumptions behind the
different theories is favoured by this approach.

xi
xii Preface
· 1 hoped that the mixture of problems covered will enable the student
I t IS a so f . . h nics and
both to obtain an overview of some important parts .o engmeenng mec a
to a reciate the wide field of applications for the fimte el~ment. method.
~he equations are numbered according to the chapter m whtch they appear, for
Suggestions to the Instructor
instance (3.27). The same system applies for references to figures and tables. References
to the literature are collected in alphabetical order at the e.nd o~ t~e book: Wh~n
· · 1s
a d efi mt10n · rna de or a concept is explained for the first ttme, ttalzc lettenng wtll
be used.

Different possible course contents are outlined below:

Chap ers 1- 11 1 minimum course

J

I
Chapters 12-16 2

J
intermediate courses
I
Chapters 17- 18 3

I
I
Chapters 19-20 4 full course

The minimum course proposed for an introduction to the finite element method
is indicated by route 1 above. It only considers problems where the unknown function
is a scalar, for instance the temperature in heat flow problems.
In addition, topics in solid mechanics are treated in the intermediate course
outlines 2 and 3. Here, Chapter 14 on Saint-Venant torsion problems may be omitted
without any adverse effect on the following chapters. However, Saint-Venant torsion
provides an interesting example of a solid mechanics problem that is analogous to
the two-dimensional heat flow problem. Course route 3 includes the finite element
formulations of beams and plates that are of great practical importance.
"' Finally, the full course indicated by route 4 also treats isoparametric finite elements

xiii
xiv Suggestions to the instructor

and numerical integration, and it should be noted that Chapters 19 and 20 may be
combined directly with the minimum course.
As an indication of the amount of time required, course propos~ls 3 and 4 ~ay
require 3- 6 effective working weeks depending on the extent of exercises and proJects
I Acknowledgements I
set by the instructor.

This textbook has grown out of years of experience of teaching the finite element
method. Many people have therefore influenced the course material in a direct or
indirect manner. In particular, we would like to express our sincere gratitude to K.-G.
Olsson, who read the entire manuscript and provided a number of valuable suggestions
and improvements. The comments given by Dr B. Bodelind and P.-E. Austrell are
also appreciated. Moreover, we are sincerely grateful to P. Nilsson and B. Zadig for
their accurate and expedient typing and drawing, respectively, which were executed
within very tight time limits.

XV
 1
[ Introduction
I
1.1 Basic description
All the physical phenomena encountered in engineering mechanics are modelled by
differential equations, and usually the problem addressed is too complicated to be
solved by classical analytical methods. The finite element method is a numerical
approach by which general differential equations can be solved in an approximate
manner ( cf. Figure 1.1 ).
The differential equation or equations, which describe the physical problem
considered, are assumed to hold over a certain region. This region may be one-,
two- or three-dimensional. It is a characteristic feature of the finite element method
that instead of seeking approximations that hold directly over the entire region, the
region is divided into smaller parts, so-called finite elements, and the approximation
is then carried out over each element. For instance, even though the variable varies
in a highly non-linear manner over the entire region, it may be a fair: approximation
to assume that the variable variesjn a linear fashion over each element. The collection
of all elements is called a finite element mesh ( cf. Figure 1.2).
When the type of approximation which is to be applied over each element has
been selected, the corresponding behaviour of each element can then be determined.
This can be performed because the approximation made over each element is fairly
simple. Having determined the behaviour of all elements, these elements are then
patched together, using some specific rules, to form the entire region, which eventually
enables us to obtain an approximate solution for the behaviour of the entire body.
The finite element (FE) method can be applied to obtain approximate solutions
for arbitrary differential equations. In {his textbook we shall only be concerned with
boundary value problems, i.e. differential equations where certain information is known
a priori for the nknowns at the boundary. For simplicity, only linear boundary
problems will be considered. However, even initial value problems can be solved by
the FE method. Initial value ~roblems are typicalfor transient phenomena like wave
propagation and transient heat conduction.
As the FE method is a numerical means of solving general differential equations,
it can be applied to various physical phenomena. In order to emphasize this aspect,

1
2 Introduction to the finite element method
Introduction 3
Fin ill'
T
Physical Modl'l Oifferl'ntial Approximation
phl'nomenon 0 equati on 0 l'll'ml'nt
equati ons

Figure 1.1 Steps in engineering mechanics analysis

Bounda ry Ell'ment
I
)> .'<:J 'A .- ~ .v ,;,. I I

ctJ
. • .p " .· . Model Approximat ion Figure 1.3 Temperature distribution along one-dimensional fin
.. <3 . •· '() '4 I I
. .' ~ . :
6. . . · 4
· , v ' . ·.
4 •
:: ~ - '4

Hl'at conduct ion Doff. equation Finite element

T2
in concrete for hl'ot con- ml'sh

I
structure duction

Figure 1.2 Illustration of modelling steps

r
Q~=~o===ot:::=:::io==39- X
I 5
we shall be concerned here with the FE formulation of such diverse problems as heat
conduction, torsion of elastic shafts, diffusion, groundwater flow, and the elastic Figure 1.4 Nodal points and temperature values at the nodal points
behaviour of one-, two- and three-dimensional bodies, including beam and plate
analysis.
T
As previously mentioned, it is a characteristic feature of the FE method that the a)
region, i.e. the body, is divided-into smalle parts, i.e. the elements, for which a rather
simple approximation is adopted. This approximation is usually a olynomial. The
approximation over each element means that an approxi mation is adopted for how
the variable changes over the element. This approximation is, in fact, some kind of
interpolation over the element, where it is assumed that the variable is known at
certain pojnts in the element. These points are called nodal points and they are often
located at the boundary of each element. T he precise manner in which the variable
changes between its values at the nodal points is expressed by the specific *~=~§===§~=~§==:Jo -----x
l 3 4 5
approximation, which may be linear, quadratic, cubic, etc.
In order to be more specific about this approximation technique, consider the true Figure 1.5 (a) Four .elements with linear temperature variation within each element;
(b) resultmg approximate temperature distribution along the fin
temperature distribution T(x) along the one-dimensional fin shown in Figure 1.3.
Assume that the temperature is known at the five positions shown in Figure 1.4.
These positions correspond to the nodal points and need not be equally spaced. Now, assume instead ~hat the fin is divided into two elements and that within
Suppose now that the fin is divided into four elements over which the temperature e~ch element a quadratic variation is adopted. This situation is shown in
is assumed to vary linearly between the nodal points. This situation is shown in Figure 1.6(a) and th.e resulting. approximate temperature distribution along the fin
then takes the form Illustrated m Figure 1.6(b ).
Figure 1.5(a) and the resulting approximate temperature distribution along the fin
is illustrated in Figure 1.5(b ). . We observe th~t irrespective of whether the linear or quadratic approximation
IS used, the approximate temperature distribution along the fin is known once the
4 Introduction to the finite element method Introducti on 5
T

.J m
a) b)
a)
cooling p1pe
T T3

ril
o
1
o
2
9 6"----=o_ _
3 4
.Jo
5
Tl

(>.__ ___;;;..o_ _....;§._____....;o'-------"9- X

2 3 5
--"" /
--
I ~Ism
6~''',\ ',
---- - \ ~'
\
\
,
',
',
---- -

',
'
J.
@
. .
Q)
steel membrane

I 75 m
Figure 1.6 (a) Two elements with quadratic temperature variation within each element; \
\
' ',
(b) resulting approximate temperature distribution along the fin \\ ' ,.
\ ..... ,
Lund
\
\
\
.'
'
"'
temperatures at the nodal points are known, i.e. these temperatures values are now
the unknowns of the problem. In this way the original problem with, in principle, 1. 10 m
infinitely many unknowns, i.e. degrees of freedom (d.o.f.), has been replaced by a
problem with a finite number of unknowns. In the present situation, the number of
unknowns is five. In general, it is obvious that the more unknowns, the more accurate b)
the approximate solution.
A system with a finite number of unknowns is called a discrete system in contrast
to the original continuous system with an infinite number of unknowns. It will turn
out later that the determination of the values of the variable at the nodal points
follows from the solution of a certain system of equations. In the case considered in
Figures 1.3- 1.6 this system of equations would consist of five equations with five
unknowns, but in general the system often involves thousands of unknowns. Obviously
such systems cannot be solved by hand and, therefore, the F E method relies entirely
on the availability of efficient computers.
As already touched upon, the FE method can be applied to arbitrary differential
equations. Moreover, arbitrary geometries of bodies consisting of arbitrary materials
can be analyzed. It is therefore no surprise that, in general, the FE method today 30°C
presents the most powerful approach for solving differential equations that occur in
engineering, physics and mathematics.
In its most basic form the FE method shares many common features with simple
matrix structural analysis. In this analysis, structures consisting of trusses and beams
are treated in a systematic manner with certain displacements as the unknowns. We
shall present such an analysis in Chapter 3. However, today's FE approach
emphasizes that arbitrary differential equations can be solved, and this much more
general viewpoint is the one we shall adopt.
0°C
The emergence of the FE method took place in the early 1960s and since then
its use has spread to virtually all fields of engineering. Some of the prominent names Figure 1.7 (a) Casting of concrete in containment vessel after replacement of nuclear
associated with the development of the FE method are Argyris, Clough and equipment; casting of region 2 occurs 72 hours after casting of region 1; (b) resulting temperature
Zienkiewicz, and for a historical account of different major contributions, we may development
17

Introduction to the finite element method Introduction 7

6
refer, for instance, to Segerlind ( 1976), Stasa ( 1985), Zienkiewicz ( 1970, 1983) and 1.2 Some specific applications
Zienkiewicz and Taylor ( 1989).
In practice, so-called general-purpose FE programs are often used. to analyze As already stressed, the FE method is applicable in widely different fields and
different problems. Even though very efficient co~puter pr?grams are avatlable today, Figures 1.7- 1.14, which show some results obtained at the University of Lund,
we want to emphasize that in order to obtam meamngful results, a profound illustrate this important fact.
understandingS>f the undedyjng theories and thei rJimitations is required by th~ u~er. A nuclear containment vessel is shown in Figure 1.7. In order to replace some
Otherwise, the FE program may produce results whic~ may look ~eat and convmcmg. of the nuclear equipment, a hole was made in the concrete containment vessel and,
when presented in beautiful colour graphics, but whtch may be melevant. w~ also after the replacement, concrete was cast to restore the containment shell. The concrete
want to underline that even though a large number of FE programs are avatlable was cast in two steps: first in region 1 within the steel membrane and 3 days later
today, research within different FE approaches is more a.ctive than eve~. This research in region 2 outside this membrane. Owing to the hydration of cement during hardening
is directed for instance, towards more efficient and rehable formulatwns as well as of the concrete, heat is developed and the resulting heat expansion may result in
towards v~rious non-linear problems. To illustrate this research activity we may.refer cracking of the concrete. To avoid such cracking, the fresh concrete in region 2 is
to Cook et a/. ( 1989), who state that 10 papers on the FE method were p~bh~hed cold, about 3°C, and, in addition, heat is removed during the hardening process by
in 1961, 134 in 1966, and 844 in 1971. By 1976, the cumulative total of FE pubhcatwns means of cooling pipes. The casting of region 2 occurred 72 hours after the casting
exceeded 7000, and by 1986, the total was about 20 000 papers. of region 1 and Figure 1.7 shows the resulting temperature development.
Figure 1.8 shows the FE mesh and the deflections of an ordinary structural
component due to lateral loading.
In Figures 1.9 and 1.10, the passenger cabin of a car is shown. The problem is
to investigate the noise level, i.e. the air pressure variation in the cabin caused by
engine vibrations. This problem constitutes a so-called coupled problem in which the
__, r -
c-)1
Hi 1--t-- I
-;I I
I
r--i I I
1--7:
--)I
I
I I
--J I
H:
_, I
I

__,: I
r--'1
1-i:
-;I
I

__,: f- l
'1: I
~I
--:)
- '·- -- I

-J I
-J
I
r I
I
I
I
I
I
I I
[--'-
~I I
I
--
t-J
-------
Figure 1.9 Finite element mesh of the structural part, i.e. the passenger cabin itself. The air
Figure 1.8 Finite element mesh and deflections of a laterally loaded structural component inside the cabin is also divided into finite elements. This cabin is exposed to engine vibrations
8 Introduction to the finite element method Introduction 9

Figure 1.12 Displacement and fi nite element mesh

steel bea m F I

Figure 1.10 · Result in terms of air pressure along the inside of the passenger cabin and
structural deformation of the cabin

Figure 1.13 (a) Steel beam supported by rubber components; (b) deformation modes of
rubber component at some resonance frequencies

A steel beam supported by rubber components is shown in Figure 1.13( a). To

investigate its dynamic response in detail, the deformation modes at some resonance
frequencies of the rubber component itself were determined (cf. Figure 1. 13(b)).
Finally, a water-filled pipe is shown in Figure 1.14(a). The deflection modes of
Figure 1.11 Wooden beam tested in laboratory the pipe at some eigenfrequencies, i.e. at resonance, are shown in Figure 1.14(b)
together with the corresponding water pressures in Figure 1.14(c). Just as in ·
structural vibrations of the passenger cabin influence the motion of the air within Figures 1.9 and 1.10, this problem constitutes a coupled problem in which the pipe
the cabin and vice versa. deflection influences the water pressure and vice versa.
Figures 1.11 and 1.12 show the behaviour of a wooden beam in which a horizontal The problems illustrated above show that widely different and complicated
crack develops due to the loading. physical phenomena can be analyzed by the FE me.t hod. Indeed, the main objective
 2
10 Introduction to the finite element method

b) c)

al [Matrix algebra
I
The finite element method is a numerical approach which results. in the establishment
of systems of equations often involving thousands of unknowns. To enable one to
deal with such expressions in a compact fashion which emphasizes the physical
content, use of matrix algebra turns out to be convenient. In this chapter we shall
present a short review of some basic matrix algebra. No proofs will be given, as the
intention is simply to recall elementary results. For a more detailed treatment the
reader is referred, for instance, to Bathe ( 1982), Hildebrand ( 1965) and Strang ( 1980).
Apart from providing a review of matrix algebra, the present chapter also serves as
an introduction to the notation adopted in this textbook.

2.1 Definitions
In general, a matrix consists of a collection of certain quantities which are termed
the components of the matrix. The components are ordered in rows and columns and
if the number of rows or columns is equal to one, the matrix is one dimensional,
otherwise it is two dimensional.
Figure 1.14 (a) Water-filled copper tube; (b) deflection modes of tube at some eigenfrequencies; As an example of a one-dimensional matrix, consider a column matrix, where the
(c) corresponding water pressures at resonance number of columns is equal to one. A column matrix is denoted usually by a lower-case
letter in bold type; for instance
of this textbook is to enable the reader to derive the FE formulation of various
problems and to appreciate the unified approach offered by this method. Befor~ this
can be achieved, some preliminary tools are necessary and, first, we shall provide a (2.1)
review of basic matrix algebra.

where c 1 , c 2 , c3 and c4 are the components of the matrix c. Apart from being one
or two dimensional, the specific dimension of a matrix is given by the number of rows
and columns, i.e. column matrix c of (2.1) has the dimension 4 x 1. The transpose

11
12 Introduction to the finite element method Matrix algebra 13

aT of a is given by the row matrix 2.2 Addition and subtraction

aT=[al a2 a3] (2.2)
If two matrices have the same dimensions they can be added and subtracted, with
which has the dimension 1 x 3. addition and subtraction being carried out for each component of the matrices. For
A two-dimensional matrix is denoted usually by an upper-case letter in bold instance, we have·
type; for instance
ell
[B" B,]
C12
B12 (2.8)
C21 c22
B = Bzl B22 B23 ; C= (2.3)
C31 c32
B31 B32 B3 3
Likewise we have
c41 c42
cT=aT ± bT=[a 1 a2 a 3]±[bl bz b3]
where B is termed a square matrix since the number of rows and columns are equal.
The dimensions of Band C are 3 x 3 and 4 x 2, respectively. The transpose BT of =[a 1 ±b1 a2 ±b2 a 3 ±b 3 ] (2.9)
B is obtained by interchanging rows and columns in B, i.e. and fo r two-dimensional matrices we have

(2.4)

and the matrix B is symmetric if B = BT. In what follows we shall often encounter
It follows that
symmetric matrices. If only the diagonal components in a matrix A are different from
zero, A is termed a diagonal matrix. As an example, A given by a ± b =±b+a
2 0 0 0 a T± bT=± bT+aT (2.11)
0 1 0 0 A±B =± B +A
A= (2.5)
0 0 - 1 0 Moreover, it appears directly that
0 0 0 3 (a± b)T=aT± bT

is a diagonal matrix. Obviously, a diagonal matrix is a square matrix. If all the (aT ± bT)T = a ± b (2.12)
diagonal components of a diagonal matrix are equal to unity, the matrix is a unit (A ± B)T = AT ± BT
matrix. A unit matrix is written as I and an example is

(2.6)
2.3 Multiplication
Any matrix can be multiplied by a number c and this implies that each component
is multiplied by the number c. Examples are
A zero matrix 0 is defined as a matrix where all components are equal to zero.
Examples are 1 1

ca = c a 2 ] = [ca
ca2 ]
O= [0oo'
OJ· O= [0ooo
0 OJ (2.7) [
a3 ca3
(2.13)
14 Introduction to the finite element method Matrix algebra 15

11 12 11 12 matrix product x T A defines the row matrix c T with the dimension 1 x n according to
A A J [ cA cA J (2.14) n
cA = c A21 A 22 = cA 21 cA 22
[ cT = xT A where ci = 2:: x ;Aii (2.22)
A31 A32 cA 31 cA32 i= 1

The scalar product of two column matrices a and b having the same dimension is In symbolic form we have
defined according to
CT = XT A (2.23)
Ix n (I x m) (m x n)
n
aTb = 2:: a;b; (2.15) An example is
i=!

where n is the number of rows. It follows that x 2 ][A11 A12 A13]

A21 A22 A23
aTb=bTa (2.16)
= [x 1A11 + x 2 A 21 x 1A 12 + x 2 A 22 x 1A 13 + x 2A23]
With a and b given by (2.1) we obtain
Finally, matrix multiplication of two-dimensional matrices can be carried out.

aT&~ [a, a,{::J ~

If A has the dimension m x n and B the dimension n x p, the matrix product AB
defines the matrix C with the dimension m x p according to
a, a,b, + a,b, + a,b, (2.17)
n
c = AB where cij = L A;k Bkj (2.24)
The length of a or aT is denoted by IaI and defined by k =1

lal = (ai +a~+···+ a;) 112 (2.18)

In symbolic form we have
where n is the number of rows in a. From (2.15) and (2.18) we have that
C = A B (2.25 )
mxp (m x n) (n x p )
lal =(aTa)
112
-1 (2.19)
As an example we have
Let us now generalize the multiplication rule (2.15) to more complicated
situations. Let A have the dimension m x n and x the dimension n x 1; then the
matrix product Ax defines the row matrix c with the dimension m x 1 according to
A12J
A22 [BB 11

21
n
A 32
c =Ax where c; = 2:: Aiixi (2.20)
- [AuBu + A12B21 AuBt2 + A12B22]
j = 1

- A21Bu + A22B21 A21 B12+ A22B22

In symbolic form we can write
A3 1B11 + A32B21 A31B12 + A32B22
c = A X (2.21)
m X I (m X n) (n X I) We emphasize that matrix multiplication is only defined if the matrices possess
As an example we obtain t he correct dimensions. Moreover, if A has the dimension m x n and B has the
dimension n x m both the multiplications AB and BA are defined, but in general

[
Au
c =Ax = A21
A31
A12J
A 22
A32
[::J =
[Aux1 + A12x2J
A21 x 1 + A 22 x 2
A31X1 + A32x2
we have
AB#BA (2.26)
This follows fro m the fact that AB defines a matrix with the dimension m x m, whereas
Likewise, let now x have the dimension m x 1 and A the dimension m x n ; then the BA defines a matrix with the dimension n x n. Moreover, even if m = n (2.26) will
16 Introduction to the finite element method Matrix algebra 17
If row number i and column number k are deleted, a new square matrix with
dimension (n - 1) x (n - 1) emerges. The determinant of this matrix is called a minor
of A and is denoted by det M ik· The cofactor of A is denoted by Aik and is defined by

(2.35)
and

BA = [Ell B 12 ][Ail A 12 ] = [B 11 A 11 + B 12 A 21 BllA 12 + B 12 A 22 ] According to the expansion formula of Laplace we then have
B 21 B 22 A2 1 A22 B21A11 + B22A21 B21A12 + B22A22
(2.28) n

det A = I A ;kAlk (2.36)

A comparison of (2.27) and (2.28) shows that (2.26) will hold unless the components k= 1

of A and B possess some specific properties.

From definition (2.24) and the definition of the unit matrix I it follows that where i indicates any row number in the range 1 :::::; i :::::; n.
AI= A (2.29) As an example consider A given by

We also recall that

(2.37 )
I (AB)T = BT AT (2.30)
It follows that
It appears that (2.30) implies
(ABC)T = ((AB)C)T = CT(AB)T = CTBT AT (2.31)

and we also have Suppose that in (2.36) we choose i = 1; then with (2.35) and (2.38) we obtain the
(Ax)T = xT AT (2.32) familiar result
2
Moreover, it follows trivially that if c denotes a number then det A = I AlkA~k = A 11 A~ 1 + A 12 A12
k=1
cAB= A(cB) (2.33)
= A 11 ( - 1) 1 + 1 det Mil + A 12 ( - 1) 1+ 2 det M 12
Finally, we recall that the distribution law from ordinary algebra also holds for
matrices; for example = A11A22 - A!2A21 (2.39)

(A + B)x = Ax + Bx The reader may verify that the same result is obtained if instead we choose i = 2 in
xT(A + B)=xTA+xTB (2.36).
(2.34) It appears that (2.36) expresses the determinant as a certain linear combination
(A + B)C = AC + BC of all components Aik in the arbitrary row i. It turns out that the expansion formula
C(A +B)= CA + CB of Laplace can also be written as a linear combination of all components in an
arbitrary column. In fact, we have

n
2.4 Determinant det A = I AkiA ~i (2.40)
k= 1

For the square matrix A, it is possible to calculate the determinant det A of A. Let
A be of dimension n x n. If n = 1 then, by definition, det A = A 11 . Before it is possible where j indicates any column number in the range 1 :::::; j :::::; n. As an example consider
to determine det A for n ;;::: 2 some definitions have to be made. A as given by (2.37) and choosej = 1. Then with (2.40), (2.35) and (2.38) we obtain
18 Introduction to the hnite element method
Matri x algebra 19
2
det A= L Ak 1 A~ 1 = A 11 A~ 1 + AztA2 t By means of the cofactor Ark of A as defined by (2.35), it is possible to present an
k= l explicit expression for the inverse matrix A- 1.
= A 11 ( -1) 1 + 1 det M 11 + A 21 ( -1)2+ 1 det M 21 For all values of i and k, we are able to construct a square matrix of dimension
n x n by means of (2.35). If this matrix is transposed, we obtain the adjoint matrix
= AttA22 - A2tA12 of A given by
in accordance with (2.39). A~~ A~2
T
A ~n
In practice, the direct use of the expansion formulae of Laplace in numerical
calculations is unsuitable since many operations are necessary. For theoretical A 2t A22 A2n
adj A = (2.45)
considerations, however, these expansion formulae are very important as they enable
one to establish a number of properties for determinants. In particular, it can be
A~t A~z A~n
shown that if:
It turns out that the inverse matrix is given by

l. a row (or column) consists of zeros, the determinant is zero; I A- 1

=adjA/ detA (2.46)
2. if two rows (or columns ) are proportional, the determinant is zero ;
3. if a row (or column) is multiplied by a factor k, the determinant is It is important to note that A -l only exists if det A# 0. A square matrix A for which
also multiplied by the factor k. det A = 0 is called a singular matrix. We also recall that
(A - l)T =( AT) - 1 (2.47)
Moreover, for later purposes it is convenient to recall that a row (or column ) operation (AB) - 1 = B - 1A- 1 (2.48)
is an operation where all components of one row (or column) are multiplied by a
and the latter expression implies
certain factor and then added to another row (or column). We then have :
(ABC) - 1 = c - 1(AB) - 1 = c - 1B - 1 A - 1 (2.49)
We note that if
4. row (or column) operations do not change the determinant ;
5. if two rows (or columns) are interchanged, the determinant changes A -l= AT (2.50)
its sign. then the matrix is called an orthogonal matrix. If A is orthogonal, we have
ATA = AAT =I.

Finally, we note that

det AT= det A ( 2.4 1)
det(AB) = det A det B (2.42)
2.6 Linear equations: number of equations is equal
to number of unknowns
whereas
det(A + B)# det A+ det B (2.43) The solution of systems of linear equations often involving thousands of unknowns
is characteristic of numerical approaches like the finite element method. It is
therefore of importance to have a firm understanding of the fundamental properties
of different types of linear equation systems. ·
2.5 Inverse matrix Consider first the system of equations

The inverse A - l of a square matrix A is defined by I Ax = b I (2.5 1)

1 1 (2.44)
A - A = AA - = I
where A is a square matrix of dimension n x n and x and b both have the dimension
20 Introduction to the finite element method Matrix algebra 21

n x 1. If b = 0 then (2.51) is called a homogeneous system of equations, otherwise it of the third row in the new coefficient matrix becomes zero. This process is continued
is termed an inhomogeneous system of equations. until all the components of the first column - except the first component - are zero.
Assume that This process is, in fact, an elimination of the x 1 -variable from all equations except the
(2.52) first. From this new system of equations, and using the second equation, the xrvariable
det A#- 0
is eliminated from the equations below the second equation. We continue this process
then by premultiplying (2.51) by A - I we obtain until the lower left part of the new coefficient matrix consists of zeros.
A-lAx= A - 1b Therefore, by this elimination technique, (2.56) is transformed into the form

which, due to (2.44), results in the solution A' x = b' (2.57)

(2.53) where we write A' and b' instead of A and b, respectively, to emphasize that the new
coefficient matrix and right-hand side have changed as a result of the elimination
It appears that the homogeneous system of equations process. In accordance with what was described above A' has the form
Ax= 0 (2.54)
A'11 A'1z A'13 A'ln
for which det A #- 0 only has the trivial solution x = 0. Therefore the non-trivial
solution of the homogeneous system of equations ( 2.54) requires that det A = 0. 0 A~z A~3 A~.
Homogeneous systems of equations are of importance in many applications, for A'= 0 0 A;3 A3. (2.58 )
instance in buckling and vibration problems.
We conclude that for homogeneous systems of equations, we have
0 0 0 A~n
Ax =0 (2.55)
where the lower left part below the diagonal consists of zeros. This form is also
• If det A = 0, a non-trivial solution exists
expressed by saying that the system of equations has been triangularized. The diagonal
• If det A #- 0, no non-trivial solution exists
elements A'11 , A~ 2 , •.. , A~. are termed pivot elements.
The system of equations (2.57) with the coefficient matrix (2.58 ) is indeed very
For inhomogeneous systems of equations, we recall that
easy to solve. From the last equation, i.e. equation number n, we obtain x. = b~ / A~•.
This solution is substituted into equation number n - 1 to provide x._ 1 and we
Ax= b; b #- 0 (2.56)
continue in this manner until all x-components have been determined. This process
• If det A #- 0, one unique solution given by (2.53) exists is called back-substitution.
• If det A = 0, no unique solution exists. Depending on the To illustrate this Gauss elimination technique, consider the system of equations
specific b-matrix we may have no solution or an infinity

- 1 ~0l[::l
of solutions -100

With these important stateme~ts in mind let us now investigate in more detail
the solution of (2.56) when det A #- 0 and where matrix A is termed the coefficient
200
- 100 100 x3
= [:l
2
(2.59)

matrix. In principle, this solution is given by (2.53 ). In practical numerical calculations,

Multiply the first equation by 1/ 2 and add to the second. This results in
however, the direct establishment of the inverse A - l is all too cumbersome and other
approaches are used. A very common and efficient approach is Gauss elimination.

-1~ol[::l = [t~l
- 100
In Gauss elimination, the equations are combined in such a manner that the
lower left part of the new coefficient matrix consists of zeros. To obtain this objective, 150
a multiple of the first equation is added to the second equation such that the first -100 100 x3 2
component of the second row in the new coefficient matrix becomes zero. A multiple
of the first equation is then added to the third equation such that the first component The first component in the third row is already zero. Therefore, multiply the second
 22 Introduction to the finite element method
Matr ix algebra 23
equation by 2/ 3 and add to the third equation to obtain
2.1 Linear equations: number of equations is different
- 100 from number of unknowns
150 (2.60)
Having treated the important case of linear equations where the number of equations
0 and unknowns is the same, we shall for later purposes investigate the properties of
The system of equations has now been triangularized in accordance with (2.58). Now homogeneous systems of equations where we have more unknowns than equations.
comes the process of back-substitution. From the third equation we obtain That is, we consider the linear homogeneous system of equations Ax = 0 where A
x 3 = 30/ 100; the second equation then provides x 2 = 28 / 100 and the first equation has the dimension m x n, whereas x has the dimension n x 1 and 0 the dimension
m x 1; moreover, n > m. The following result is obtained:
results in x 1 = 18/100, i.e.

A x = 0 (2.65)
(m x n) (n xl) mxl

• For n > m, i.e. more unknowns than equations, we have at least

Having illustrated the Gauss elimination technique, we return to the general n - m non-trivial solutions
triangularized form of (2.57) and (2.58). Let us calculate the determinant det A' of
the coefficient matrix (2.58). To do this it is convenient to use the expansion formula
of Laplace given by (2.40) where the expansion takes place along some column. In
the present case we choose the first column, i.e.j = 1. As the only non-zero component
in this column is A'11 we obtain from (2.40)
n n
2.8 Quadratic forms and positive definiteness
det A'= L A~iA~j = L A~~A~cl = A '11A't1
k=l k=l
(2.61)
For a square matrix A consider the quantity xT Ax, which is a number. This quantity
is called a quadratic form. If
From (2.35) the cofactor A~\ is
A '1°1 = det M 11 (2.62) x TAx > 0 for all x =f. 0 (2.66)
To determine the minor det M 11 we delete the first row and first column in the
A'-matrix and calculate the determinant of the remaining matrix. It appears readily then the matrix A is said to be positive definite. In wha t follows, we shall often
that steps identical with (2.61) and (2.62) will result. Therefore, the reader may easily encounter positi ve definite matrices.
verify that we end up with the simple result It follows directly from (2.66) that the determinant of a positive definite matrix
(2.63) is different from zero. To show this assume that (2.66) holds and that det A = 0.
Since det A = 0, it is possible to determine a non-trivial solution to the homogeneous
i.e. the determinant is the product of all the pivot elements. system of equations Ax = 0, which means that (2.66) is violated. Therefore det A =f. 0
As the determinant of A' turns out to be very easy to identify, it is of interest to must hold for positive definite matrices, i.e.
investigate how this determinant relates to the determinant of the original coefficient
matrix A of (2.56). For this purpose we observe that A' is obtained from A by making
If A is positive definite then det A =f. 0 (2.67)
suitable row operations. Referring to statement ( 4) on page 18 we immediately
conclude that
Note that the inverse argument does not hold, i.e. if det A =f. 0 we cannot conclude
I det A'= det A (2.64) that A is positive definite.
Finally, if a matrix is positive definite it is required that all the diagonal
and the reader may check this conclusion by using (2.59) and (2.60). Often the components of the matrix are positive. To show this, choose x as
establishment of the determinant is most easily based on (2.63) and (2.64). x T = [0 ... 0 xk 0 ... OJ

L
24 Introduction to the finite element method Matrix algebra 25

then it follows from (2.66) that Define the matrices y, z, g and h by

xTAx= Akkx~ > 0

which proves the statement above.
Y= [::l z = [x 3 ]; g = [~:l h = [f3 ] (2.74)

We also mention that the square matrix A is called positive semi-definite if Then the system of equations (2.72) may be written as

x TAx ;:::: 0 for all x '# 0 (2.68) (2.75)

Multiplication of the matrices implies

By+ Cz =g
(2.76)
Dy + Ez = h
2.9 Partitioning By inspection, the reader may check that (2.72) and (2.76) are, in fact, identical. The
important point to observe is that the partitioning must be made in such a manner
To facilitate matrix manipulations, it may be useful to partmon a matrix into that the multiplications in (2.76) are meaningful, i.e. that the submatrices possess the
submatrices. A submatrix is a matrix that is obtained from the original matrix by correct dimensions. By partitioning matrices, it follows that not only multiplications,
including only the components of certain rows and columns. The idea is demonstrated but also additions and subtractions can be carried out.
below, where the dashed lines are the lines of partitioning:

(2.69) 2.10 Differentiation and integration

Consider the matrix A given by

Using this partitioning the matrix A may be written as (2.77)

A = [~ ~] (2.70) If the components of this matrix depend on a variable x, we define matrix differentiation
as
where
dA-11 12 dA 13]
d A-
- - --
dA dx dx dx
B = [A11 A12]; C = [A13] (2.78)
A21 A zz Az3 (2.71)
dx = dA 21 dA 22 dA 23
[ -- -- - -
D = [A31 A32]; E = [A33] dx dx dx
As an example of the use of partitioning, consider the system of equations i.e. all components are differentiated with respect to x. As an example, consider the
system of equations
Ax = f (2.72)
A(x)b = f(x) (2.79)
where A is given by (2.69) and x and fare given by
where A and f depend on x, whereas b is constant. Differentiation of (2.79) yields

dAb = df (2.80)
(2.73) dx dx
If the matrix multiplication in (2.79) is carried out and if differentiation is then
26 Introduction to the finite element method

performed, it is easily shown that the result is identic~! wi.th (2.8?). .we conclude that
3
the usual rules for differentiation hold even for matnx differentiatiOns.
In a similar manner, integration ofthe matrix A given by (2.77) is defined as an Direct approach- truss
integration of each component, i.e.
analysis
I
I = [I A11 dx I A12 dx IA 13 dx]
(2.81)
Adx
A 11 dx I A 21 d x IA 13 dx

As already touched upon in the introduction, the finite element method is based on
the concept that the body in question is divided into regions, so-called elements, for
which the behaviour can be described in a simpler manner than the behaviour of the
entire body. When the behaviour of the elements has been formulated, these elements
are patched together according to some rules, thereby enabling one to predict the
behaviour of the entire body.
We shall discuss later how the behaviour of the elements themselves can be
obtained for general situations using the finite element method, but in some simple
cases this behaviour can be obtained directly. The present chapter is devoted to a
study of such a simple situation, namely the response of elastic springs which may
form a conventional truss structure. We emphasize that apart from the manner in
which the element characteristics are established, the remaining steps in the calculation
are the same as for the general finite element method. Therefore, this chapter serves
as a prelude to many of the features that we will encounter in the following chapters.

3.1 Simple structure consisting of springs

We shall present a systematic solution to the response of a structure consistmg
of interacting elastic springs. For this purpose, consider the structure shown in
Figure 3.1, where two elastic springs with stiffnesses k 1 and k2 are connected in series
along the x-axis and where the loading consists of three external forces F 1 , F 2 and
F 3 . These forces are directed along the springs. The positive direction of fo rces is
chosen to be in the direction of the x-axis.
In order to analyze the entire structure in a systematic manner, which can easily
be applied to more complicated spring systems, we split the structure into smaller
parts, i.e. elements, for which we directly know the response. The ends of each element
are called nodal points. In the present situa tion it is natural to split the structure
into two elements corresponding to three nodal points as shown in Figure 3.2.
The structure in Figure 3.1 can now be illustrated as shown in Figure 3.3.

27
28 Introduction to the finite element method Direct approach - truss analysis 29
1 k 2
-P1 ~- Pz ,______.X
...___x
u2
Figure 3.1 Two springs loaded by external forces
Figure 3.5 Characterization of one element

Element CD Element®
k1 k2 Instead of the description of the element shown by Figure 3.4 we shall adopt the
one indicated in Figure 3.5, where the end points as well as the forces P 1 and P 2
Nodal point 1 Nodal point 2 Nodal point 3 and the displacements u 1 and u2 at the end points are identified. We emphasize that
P 1 and P 2 are the element forces , i.e. forces acting on the element, and these forces
Figure 3.2 Division of structure into elements may be different from the external forces acting on the nodal points of Figure 3.3.
Likewise, one should note that u 1 and u 2 of Figure 3.5 are element displacements, i.e.
F1 F2 F3 displacements at the end points of the element. By comparison of Figures 3.4 and
CD 0 3.5, we have P 2 = N, i.e. we obtain from (3.1) that
>----X

u1
k1 2
Uz
kz 3
--
u3
P 2 = k(u 2 - ud
Moreover, it follows that P 1 =- N, i.e.
(3.2)

Figure 3.3 Structure characterized by forces, displacements, elements and nodal points
P 1 = k(u 1 - u2 ) (3.3)
We then observe that
1 k 2
N- ~ -N ,______.X
p 1 + p2 =0 (3.4)
u1 uz
This equation is an example of a balance equation, which in the present situation is
Figure 3.4 Force N within an element an equilibrium equation expressing the fact that the sum of all forces acting on the
element is equal to zero.
In this figure the displacements u 1 , u 2 and u 3 of the nodal points are also indicated. We can combine (3.2) and (3.3) into
In order to be systematic, the positive direction of the displacements is the same as
the positive direction of the forces, i.e. in the direction of the x-axis. The external
forces F u F 2 and F 3 are assumed to be applied at the nodal points. It appears that,
by means of F igure 3.3, we have obtained a systematic description of the structure or
considered.
Let us now evaluate the response of an element, i.e. one spring, and begin by K•a• = f• (3.5)
recalling some elementary facts. For this purpose, consider first Figure 3.4, where an
element in equilibrium is loaded by two oppositely directed forces. The force N within where
the element is given by
(3.1) K• = [ -~ -~} a• = [::J r· = [;:] (3.6)

where u 1 and u 2 are the displacements of the end points of the element. These K• is termed the element stiffness matrix, a• the nodal displacement vector for the
displacements are positive when they are directed in the direction of the x-axis. element and f• the element force vector. We observe that K• is symmetric.
Moreover, it appears that when the spring is exposed to a tensile force, N is positive. Equation (3.5) is the element stiffness relation and it describes the relation between
Expression (3.1) is an example of a constitutive law, i.e. a relation which describes element displacements and element forces at the end points.
how the material behaves. In the present case we have a linear elastic response in Let us now split the structure shown in Figure 3.3 so that the response of each
accordance with Hooke's law. element becomes apparent. This is shown in Figure 3.6 where urp
refers to the element
30 Introduction to the finite element method D irect approach - truss analysi s 31

structure, it is convenient to rewrite (3.1 1) and (3.12) to obtain the forms

~X

2 3
( 3.13)
u, Uz

<D and
P
-2 ..__X

uCD
1

Figure 3.6
uCD
2

Identification of structure and splitting into elements

[~
0 -k 2
:2 _:2][::] [p:]
k2 u3
=
P2
( 3.14)

The relations ( 3.13) and ( 3.14) are termed the expanded element relations since they
comprise the element relations written in an expanded form- so that they involve the
displacement at end point 1 of element 1 whereas, for instance, PfY refers to the
unknowns u 1 , u 2 and u3 of the entire structure. We may write (3.13) applicable for
element force at end point 1 of element 2.
element 1 as
With this notation and referring to (3.5), we obtain for element 1

[ k1 -k1J[uPJ
-k 1
[pep]
ug:; - pg:;
k1
(3.7)
where
K~·a = f~• (3.15)

and, likewise, we obtain for element 2

k2
[ -k
2
-k J[uPJ
2
uP
k2
[PPJ = PCf!
(3.8) (3.16)

Equations (3.7) and (3.8) describe the behaviour of two elements independently of
K~· is termed the expanded element stiffness matrix for element 1, a the nodal
each other without recognizing that the structure comprises these elements. In reality
the displacements u 1 , u 2 , u 3 of the structure are related to the element displacements, displacement vector for the entire structure and f1• the expanded element force vector
uP, ug:; and uP, uP. From Figure 3.6 appears that for element 1. We observe that K1• is symmetric. Likewise, for element 2, we obtain
from (3.14)
U CD
1 -- u 1>· u<ZI
2 -- u 3 (3.9)
K~•a = f~• (3.17)
Moreover, the two springs are connected at nodal point 2, i.e.
ug:; = uCfJ = u2 (3.10) where
With (3.9) and (3.10), equations (3.7) and (3.8) become

K~' ~ [~ :,Jr,· ~ [p:]

0
kz (3.18)
( 3.11)
- kz kz Pz
and In the establishment of the element stiffness relation ( 3.5) we have used the

k2
[ - k
2
-k J[Uz] [PPJ
2
k2 u3 = PCf!
(3.12)
equilibrium condition for each element ( cf. (3.4 )). The element forces P 1 and P 2 acting
on the element and shown in Figure 3.5 enter this element stiffness relation. For the
structure in question, all element forces are shown in Figure 3.6. It is obvious, however,
The conditions (3.9) and (3.10) express the fact that the elements are connected or that these element forces are related in one way or another to the external forces
compatible, i.e. (3.9) and (3.10) express conditions for displacement compatibility. F 1 , F 2 , F 3 acting on the structure. This relation can be derived by expressing the
As we have three unknowns in terms of the displacements u 1 , u2 and u 3 of the fact that each nodal point of the entire structure is in equilibrium. The external forces
32 Introduction to the finite el ement m e thod Di r ect approach - tr uss analysis 33

al the response of the entire structure. From (3.15), (3.17) and (3.23) it follows that

2 3 K~·a + K2•a = f (3.24)

i.e.
b)

Ka = f (3.25)
3
cl
/' F
---..... ', /,.. F
---.... ' , ---, ',
/ ,... F
where
3
/ , \A CD B/ ..2._ \C ® D/ \
2
o-~\____.,...~___..1.......__-o-- ; -~- ~ ...,._~
I
\ 1 ) p<D p(D \ 2 pQ) pQ) \ 3 /
I
K = K~· + Ki• or K = L K••
a (3.26)
' , .... ___., / 1 2 \ .......... __ ., /
I
1 2 \' .... / /
I a= l

Figure 3.7 Forces acting on each nodal point Written explicitly (3.25) takes the form

F 1 , F 2 , F 3 on the structure are shown again in Figure 3. 7 (a). Let us now cut the (3.27)
structure at points A, B, C and D as shown in Figure 3.7(b) so that the structure is
divided into free bodies. Using the principle of action and reaction, all the forces on
all the nodal points are illustrated in Figure 3.7(c). In (3.25) K is the total or global stiffness matrix of the structure, the column matrix
Equilibrium for nodal point 1 requires that F 1 - PC[> = 0, i.e. a contains the displacements of the nodal points of the structure and the column
matrix f contains the external forces applied at the nodal points of the structure. We
F! = PP (3.19) observe from (3.27) that K is symmetric. A system of equations like (3.25) is typical
Likewise equilibrium for nodal points 2 and 3 requires that of finite element calculations and the symmetry of K is also typical. The process by
which the system of equations (3.25) is constructed from the response of each element
F2 = PCjJ + P~; F3 = PP (3.20) for the entire structure (cf. (3.23) and (3.26)) is called assembling. In this assembling
respectively. Equations (3.19) and (3.20) can be combined into process we have used conditions for compatibility, namely that the elements are
connected, as well as equilibrium conditions for the nodal points. We shall find that
the relations (3.23) and (3.26) used in the assembling process are characteristic of
( 3.21) general finite element formulations.
We emphasize that the element stiffness relation was based on a constitutive
relation - Hooke ' s law, equation (3.1) - and the condition of equilibrium for the
The convention that the positive directions for element forces and external forces are element ( cf. (3.4 )). The assembling process, by which the system of equations (3.25)
the same is responsible for this simple result. Let us define the force vector f by for the entire structure is constructed, is based on compatibility (i.e. the elements are
connected, cf. (3.9) and (3.10)) as well as on requiring equilibrium for each nodal
point (cf. (3.21) or (3.23)). Equation (3.23) is remarkable as it states that the external
(3.22) force at a nodal point is equal to the sum of all element forces related to that nodal
point. This simple result follows from equilibrium - the principle of action and reaction
- and from the fact that we have chosen the same positive direction for exte'fnal
Use of(3.16), (3.18) and (3.22) enables us to write (3.21) as forces and element forces. F rom (3.23) it follows that we can just add the expanded
element stiffness matrices given by (3.16) and (3.18) to obtain the total stiffness matrix
r~· + fi• = f I or f = I
a= !
r~· ( 3.23) K (cf. (3.26)). Once these points are fully understood by the reader, it is in fact very
easy to construct the equation system (3.25) directly without going through all
the detailed steps given above.
We are now in a position to establish the system of equations that determines If all the displacements a are known, the forces f which must be applied in order
34 Introduction to the finite element method Direct approach - truss anal ysis 35
to obtain these displacements can be found directly from (3.25). In practice, the We have already mentioned that due to the enforcement of the boundary
objective is most often to determine the displacements a when (3.25) is given. However, conditions the determinant of the stiffness matrix appearing in (3.30) is different from
to solve (3.25) we know from Chapter 2 that the determinant of K is essential. From zero. It is of interest to investigate whether this symmetric stiffness matrix is positive
(3.25) and (3.27) it is easily shown that definite. Let x be an arbitrary column matrix; then we obtain

(3.28)
det K =0
(3.32)
i.e. the system of equations (3.25) possesses no unique solution. It is import~nt. to
understand the physical evaluation of this mathematical statemen~. ~n est~~hs~ng
(3.25) we have, among other things, used the fact that the structure ISm eqmhbnum. i.e. this stiffness matrix is certainly positive definite (cf. (2.66)). The fact that the
Indeed, by adding the three equations of ( 3.27) it ~ollows that F 1 + F 2 + F ~ = 0, stiffness matrix of ( 3.30) is symmetric, positive definite and has a determinant different
which is the equilibrium condition for the ent1re structure. However, 1f no from zero is characteristic of finite element applications.
displacements are prescribed a priori, the structure is in. equilibrium f~r infinitely The systematic approach described above may seem somewhat cumbersome, but
many positions; it may even move with a constant veloct~Y· Ther~f~re, m orde~ to it lends itself directly to a computerized solution, where the stiffness relation for each
obtain a unique solution for the displacements we must spectfy that ng1d-body mot1~ns element and the specification of how the elements are connected are the only
are not allowed. This can be achieved by specifying at least one nodal pomt information required in order to establish the system of equations governing the
response of the entire structure. The enforcement of the boundary conditions then
displacement. .
Let us, for example, specify that u 1 = 0. Then (3.27) can be wntten as enables one to solve this system of equations.
The approach adopted for the spring system is an example of matrix structural
(3.29)
analysis. Moreover, it is a so-called displacement method since the displacements are
considered as the unknowns. However, apart from the establishment of the element
and
stiffness relation, which could be constructed directly here, the remaining steps in the
(3.30)
analysis turn out to be identical with those of the general finite element (FE) method.
We can therefore summarize the following basic steps:
The determinant of this coefficient matrix is k 1 k 1 , which is different from zero, i.e.
(3.30) possesses a unique solution given by
Steps in the FE method
u1 = : (F 2 + F 3); u3 = :~ + (:1
+ L)F3 (3.31)
1. Establishment of stiffness relations for each element. Material
1
properties and equilibrium conditions for each element are used in
Assume that the external forces F 2 and F 3 are given; then the displacements u 2 and
this establishment.
u are determined from (3.31 ). Knowing the u 2 -value, the external force F 1 is
2. Enforcement of compatibility, i.e. the elements are connected.
d~termined from ( 3.29 ). We observe that ifF 2 and F 3 are positive then F 1 is negative,
3. Enforcement of equilibrium conditions for the whole structure, in the
i.e. the external force F 1 is directed in the negative direction of the x-axis. In fact, F 1
present case for the nodal points.
is the reaction force at nodal point 1 which must be applied in order to keep the
4. By means of 2. and 3. the system of equations is constructed for the
structure in equilibrium. From (3.29) and (3.30) it is easy to convince oneself that
whole structure. This step is called assembling.
F + F + F = 0 holds, as required for equilibrium. When ~11 the displ~cements u 1 ,
1 2 3 5. In order to solve the system of equations for the whole structure, the
u and u are known, the force N in each element can be determmed according to (3.1 ).
3 boundary conditions are enforced.
The specification that u 1 = 0 is an example of a boundary c?~diti?n, and we
2
6. Solution of the system of equations.
emphasize that the enforcement of at least one su~h boundary ~ond1t~on IS necessary
in order to solve the system of equations, otherwtse the determmant 1s equal to zero
(cf. (3.28)). Moreover, when u 1 is specified then the reaction force F 1 is initially
unknown. In contrast, when F 2 and F 3 are specified, the displacements U2 and u3 Here, a very detailed derivation of the steps has been adopted, but once these steps
are initially unknown. More generally, for a given structure it is not possible to have been understood, the reader may verify that it is rather straightforward to
prescribe the force and the displacement at the same position. establish the system of equations for the whole structure.
36 Introduction to the finite el ement method Direct approach - truss analysis 37

Let us emphasize the following points: banded, i.e. non-zero terms exist only about the diagonal. This is a common feature
of FE equations.
1. All elements are located along the x-axis. For each element, and referring to
Figure 3.5, end point 1 is always located to the left of end point 2, i.e. no ambiguity
exists as to how the element is oriented. We also note that these end points are
often called local or element nodal points as distinguished from the (global) nodal 3.2 Truss analysis -local and global coordinates
points of the entire structure. Since the elements only can be directed in one way,
it follows that for a given global nodal point the external force applied at that The spring element can even be applied to the important engineering problem of
nodal point is always equal to the sum of all the element forces related to that structures consisting of trusses, i.e. bars joined by frictionless hinges which imply that
nodal point. This observation is a consequence of equilibrium as well as of the only tension or compression exists in the bars. A typical t russ structure is shown in
systematic orientation of forces. An example is given by (3.21). Figure 3.9 and apart from its engineering importance the analysis of truss structures
2. This important point, in combination with the fact that the elements are joined provides us with the opportunity to discuss the concept of local and global coordinate
together to form the structure, i.e. the compatibility requirement, imply the systems.
essential assembling procedure which establishes the global stiffness matrix K as With reference to the spring element, Figure 3.4, let us first establish the stiffness
the sum of the expanded element stiffnesses and the right-hand side f of (3.25) k for a bar loaded by the forces N as shown in Figure 3.1 0. The bar has length L
as the external forces at the nodal points. and its cross-section is given by A.
To illustrate that once the steps above are understood one may directly establish By definition, t he force N is related to the normal stress a by
the system of equations which controls the behaviour of the entire structure, let us N = Aa ( 3.34)
consider Figure 3.8. In this structure, five nodal points exist where the external forces
F 1 ... F 5 may be applied. The bar is assumed to behave as linearly elastic according to Hooke's law, i.e.
The displacements of the nodal points are denoted by u 1 ... u 5 • Moreover, the a= Ee (3.35)
stiffnesses of the five springs are given by k 1 ... k 5 . The reader may verify that t he
following system of equations can be written down directly: where E is Young's modulus and e is the normal strain. By definition, e is determined as

2 3 4 5 (3.36)
L
1 kl -kl 0 0 0 ul Ft
where u 1 and u 2 are the displacements of end points 1 and 2, respectively. The positive
2 -kl kl + k2 + k4 - k2 -k4 0 u2 F2
direction of these displacements is in the direction of the x-axis ( cf. Figure 3.10).
3 0 -k2 k2 + k3 -k3 0 u3 F3
4 0 -k4 -k3 k3 + k 4 + k 5 - ks u4 F4
5 0 0 0 -ks ks Us Fs

( 3.33)

We recall that before this system of equations can be solved, the boundary conditions
Figure 3.9 Two-dimensional truss structure consisting of bars joined by frictionless hinges
must be enforced. We also observe that the stiffness matrices of (3.33) and (3.27) are

u, E. A
>--X
N- - if 2
-N >---X

L
I·
Figure 3.8 Structure consisting of five springs Figure 3.10 Loading of bar
38 Introduction to the finite element method
Direct approach - truss analysis 39
y
y

~-------------------------------x
~-------------------------------x
Figure 3.11 Displacement and force components of bar element in global xy-coordinate system
Figure 3.12 Displacement and force components of bar element in local x ' y' -coordinate system

Combining (3.34)- (3.36) we obtain we have the element forces P3 and P~ as well as the element displacements u3 and
u~. The positive directions of these forces and displacements are in the direction of
N = k(u 2 - U1 ) ( 3.37)
the local coordinate axes, as shown in Fig~Jre 3.12.
where the stiffness k is given by From Figure 3.5 and equations (3.5) and (3.6) we obtain

k = AE/L I (3.38) (3.39)

It appears that when k is determined according to (3.38), the spring element can be where the stiffness k is determined from (3.38). Equation (3.39) can be rewritten in
applied to a bar oriented along the x-axis. the form
Now the important point is that the structure shown in Figure 3.9 is two
dimensional, whereas our previous spring analysis was confined to springs joined Ke' a•' = f•' (3.40)
along the x-axis. To enable us to analyze a structure like that shown in Figure 3.9,
we consider a bar element located in the xy-coordinate system as shown in where
Figure 3.11. This element has the global nodal points i and j as shown.
At each global nodal point we may have element forces and element displacements u'1 P'1
with components in the x- and y-directions as shown in Figure 3.11. At nodal point u2_ P'2
i, we have the element forces P; and P;+ 1 as well as the element displacements u; a
e'
= r·· = ( 3.41)
u'3 P3
and u; +1 . Likewise at nodal point j, we have the element forces Pi and Pi+ 1 as well
as the element displacements ui and ui+1 . The positive directions of these forces and u~ P'4
displacements are in the direction of the coordinate axes as shown in the figure. Our and
objective is to determine the relation between the forces P;, P;+ 1 , Pi, Pi+ 1 and the
displacements u;, ui +l • ui, ui+l · 0 -1 0
F or this purpose, and before any loading is applied, we introduce a local 0 0 0 0
x ' y'-coordinate system with origin at end point 1 and with the x'-axis directed along Kc' = k (3.42)
the bar axis from end point 1 towards end point 2. Moreover, the y'-axis is chosen - 1 0 0
such that the x' y' -system can be obtained by a rotation of the xy-system. This rotation 0 0 0 0
is given by the angle ¢. We then arrive at the situation shown in Figure 3.12.
In this figure we also have element forces and element displacements at each end We observe that displacements along the y'-axis, i.e. u2. and u~, create no fo rces in
point, but now these forces and displacements are given in accordance with their the bar. This is a consequence of our assumption that the displacements are small.
components in the x'- and y'-directions. At end point 1 we have the element forces If large displacements were allowed, the components u2. and u~ might result in an
P~ and P2. as well as the element displacements u'1 and u2_. Likewise at end point 2,
elongation of the bar and thus in the development of forces. Moreover, the so-called
local element stiffness matrix K• ' turns out to be symmetric.
40 Introduction to the finite elemen t method Direct approach - truss analysis 41

Let us now establish a relation between the displacements u;, u;+ 1 , ui , ui+ 1 and displacements. Similarly with (3.45) and (3.46) we obtain that
u'1 ... u~. As shown in Figure 3.12, the x '-axis makes an angle cjJ with the global x-axis.
By geometrical arguments, it follows directly from Figures 3.11 and 3.12 that I f•=LTf•' (3.51)

u; = u '1 cos cjJ- u~ sin cp; ui + l = u '1 sin cjJ + u~ cos cjJ (3.43)
where
Likewise, we obtain
P'1
ui = u3 cos cjJ - u~ sin cp; ui+ 1 = u'3 sin cjJ + u~ cos cjJ (3.44)
p~
Equations (3.43) and (3.44) can be combined into r· = r·· = (3.52)
P3
( 3.45) p~

With these preliminary remarks, we are now in a position to determine the

where
relation between the element forces f• and the element displacements a•. Insertion of
U; u'1 (3.50) into (3.40) yields
U; + 1 u~
a• = a•' = (3.46)
ui u3 and premultiplication by LT and use of (3.51) result in
u '4

and
ui+ I
I K•a• = r· (3.53)

cos cjJ -sin cjJ 0 0 where the element stiffness matrix K• is given by

sin cjJ cos cjJ 0 0

LT = (3.47) (3.54)
0 0 cos cjJ - sin cjJ
sin cjJ cos cjJ We observe that once the local element stiffness relation (3.42) is known the global
0 0
element stiffness relation ( 3.53) can be obtained by means of the transformation
where L denotes the so-called transformation matrix. This matrix possesses a matrix L .
remarkable property. With U given by (3.47) it can be confirmed that the following It may be of interest to obtain an explicit expression for the global element
relation holds: stiffness matrix K• from (3.54) with K•' given by (3.42) and V given by (3.47). Simple
calculations show that
I LTL=LLT=I (3.48)
cos 2 cjJ cos cjJ sin cjJ - cos 2 cjJ - cos cjJ sin cjJ
The transformation matrix Lis therefore an orthogonal matrix (cf. (2.44) and (2.50)) cos cjJ sin cjJ sin 2 cjJ - cos cjJ sin cjJ - sin 2 cjJ
and we have that K• = k ( 3.55)
-cos 2 cjJ - cos cjJ sin cjJ cos 2 cjJ cos cjJ sin cjJ
(3.49) - cos cjJ sin cjJ - sin 2 cjJ cos cjJ sin cjJ sin 2 cjJ

Premultiplying (3.45) by Land using relation (3.48), we then conclude that and it appears that the global element stiffness matrix K• is symmetric.
One question may be raised: namely, how do we choose the local end point 1
(3.50) and local end point 2? To illustrate this problem, consider again the bar element
shown in Figure 3.11. When choosing the local coordinate system we then have the
As both forces and displacements are vector quantities, the element forces P;, two possibilities shown in Figure 3.13.
P'1 ... P~ are related in exactly the same manner as the element
P; + 1 , Pi, Pi + 1 and In Figure 3.13(a) we have chosen the same possibility as shown in Fi_gure 3.12,
42 Introduction to the finite element method Direct approach - tr uss analysis 43
y y
a) b)

~~A .
0x' y
y
2

~--------------------x

Figure 3.13
~--------------------x

Two possibilities for choosing the local coordinate system

L.
y 3

L. /D------------~
"' E. A
r2
Figure 3.15 Forces and displacements acting on the elements

With reference to (3.38) and (3.55) and using¢= 0 for element 1 we obtain for
this element

~I 1 0 -1 0 0 0
L
0 0 0 0 0 0
Figure 3.14 Simple truss structure
AE - 1 0 1 0 0 0 u3
L 0 0 0 0 0 0 u4
or K~·a = f~• ( 3.56)
but the alternative choice shown in Figure 3.13 (b) might also be taken. In 0 0 0 0 0 0 u5
Figure 3.13(a) the local coordinate system has been obtained from the global
0 0 0 0 0 0 u6
coordinate system by a rotation angle¢, whereas Figure 3.1 3(b) requires the rotation
angle¢ + 180°.
With the choice shown in Figure 3.13(a) we derived the result given by (3.53),
For element 2 the length is L J2 and ¢ = 135°. As cos 135° = -1 1J2 and
sin 135° = 1/ .fi, we obtain
where the global element stiffness matrix K• is given by (3.55). It appears that the
only information related to the choice of the local x' y' -coordinate system is the angle 0 0 0 0 0 0
¢and it also appears directly from (3.55) that if¢ is replaced by¢+ 180°, the global 0 0 0 0 0 0
element stiffness matrix K• remains the same. This means that the two possible choices
of the local coordinate system shown in Figure 3.13 result in the same global stiffness
AE 0 0 l.
2 -t -t t
Lj2 (3.57)
matrix. 0 0 -t .1
2
.1.
2 -t
As an example of the use of the element stiffness matrix, consider the simple truss 0 0 -! l.
2
l.
2 -t
structure shown in Figure 3.14.
The bars have the same £-modulus and cross-sectional area A. For clarity let
0 0 .l
2 -! -t .1.
2

us split the structure into two elements and also identify all displacements and fo rces As already indicated, expressions (3.56) and (3.57) are in fact the expanded element
acting on each element. This is shown in Figure 3.15, where we have already made relations since each element is described by the nodal displacements for the entire
use of the fact that the elements are joined to form the structure, i.e. the displacements structure. Let us now consider all the external forces that may be applied to the
shown in Figure 3.15 are the displacements of the giobal nodal points. structure. These external forces are shown in Figure 3.16, and we note that they are
44 Introduction to the finite element method Direct approach - truss analysis 45
•
considered as positive quantities when they are in the direction of the coordinate
axes, just like the element forces.
We emphasize that these external forces act on the nodal points, whereas the
forces shown in Figure 3.15 act on the elements. Using the principle of action and
y
reaction and considering the equilibrium for each nodal point, we obtain from
Figures 3.15 and 3.16 that

F3 =P 3 +P 5 ; F4 =P 4 +P6 (3.58)
L.
F5 =P1; F6 =Ps
Figure 3. 16 External fo rces applied to the structure
These expressions can be combined into
P1 0 F1 We observe that, due to the systematic choice of direction of forces, any external
Pz 0 Fz force applied to a nodal point will always be equal to the sum of those element forces
joining that nodal point. Examples are given by (3.21 ) and ( 3.59). Once this important
p3 Ps F3
+ p6 or r~· + r~· = r (3.59) point is understood, it suffices to consider only the expanded element stiffness matrices
p4 F4 and then to add all these matrices to obtain the global stiffness matrix K ( cf. also
0 p1 Fs (3.26)). The right-hand side f of the total system of equations is always equal to the
external forces and we need not show this for every application, because we have
0 Pa F6 now proved that this is a general statement.
It appears that these equilibrium conditions are similar to t~ose obtained fro~ In (3.60) we have not enforced the actual boundary conditions fo r the structure.
Figure 3.7 and equation (3.21). Adding (3.56) and (3.57) and usmg (3.59) we obtam Referring to Figures 3.14 and 3.15, we know that the displacements at the supports
are zero, i.e.
1 0 -1 0 0 0 ul Fl
(3.62)

0 0 0 0 0 0 Uz Fz At nodal point 2 we do not know the displacements u 3 and u4 , but instead we know
the external forces. Referring to Figures 3.14 and 3.16 we have
1 1 1 1
-1 0 1 + -- - -- u3 F3
(3.63)
AE 2~ 2~ 2~ 2~
(3.60)
L 1 1 1 1 Using (3.62) and (3.63) in (3.60) we obtain
0 0 --- u4 F4
2~ 2~ 2~ 2~
1 1 1 1 (3.64)
0 0 --- Us Fs
2~ 2~ 2~ 2~ (3.65)
1 1 1 1
0 0 --- u6 F6 and
2~ 2~ 2~ 2~
which is the system of equations which controls the behaviour of the structure. In
an obvious notation we write (3.66)

Ka = f where K = Kj• + K~· (3.61)

 •

Introduction to the finite element method Direct approach - truss analysis 47

46
least one element. The reader is invited to check these general observations with the
as well as
stiffness matrix given by ( 3.60).
(3.67) This discussion completes the description of the discrete approach. We emphasize
that apart from the element stiffness relation, which could be established here directly,
the remaining steps are identical to those of the general FE method ( cf. the steps on
The solution of (3.66) yields page 35 ). The critical point in the discrete approach was the assembling process,
FL r-; FL which was treated here in great detail. In the general FE method, however, this
u3 =-- ; u4 =- (1 + 2v 2) - (3.68) assembling process is taken care of automatically, but it is important that the reader
AE AE
is able to understand the physics behind this assembling scheme. Indeed, this is one
and from this solution the external forces F 1 , F 2 , F 5 and F 6 , which are in fact the of the major objectives of this chapter. For further information on the topics dealt
reaction forces from the supports, can be determined. We obtain with, reference is made to Cook et al. (1989), Przemieniecki (1968), Stasa ( 1985),
(3.69) Thelandersson ( 1984) and Zienkiewicz and Taylor (1989).

Finally, let us evaluate the physical meaning of the stiffness matrix K. In general,
we obtain an equation in the form of (3.61 ). Assume that the nodal displacements a
are given by
u1 0
Uz 0

a= ( 3.70)
ul 1

un 0
where n is the number of degrees of freedom (d.o.f.) of the structure. In (3.70) all
nodal displacements are equal to zero except u1 = 1. Using this a-matrix in (3.61)
results in
K li !1
K 21 !2
(3.71 )
K li jj

Knj fn
Therefore, when the loading of the structure is such that all displacements are kept
at zero except u . = 1, this loading implies that the external forces fare given by ( 3.71)
That is, the col~mn j of K represents the external nodal forces associated with the
activation of the unit displacement u1 = 1. However, we note that activation of one
d.o.f. creates nodal forces in only that element or elements which contain the d.o.f. in
question. Other elements are not strained and produce no nodal forces. As component
K .. of K is the nodal force in node i created by a unit displacement of u1, we conclude
th,~t K IJ.. is zero unless the degrees of freedom given by i and j are both present in at
 4 _ _ _ _ ____ a) T
Strong and weak forms - 1-D heat flow

b)
49

Strong and weak

formulations - one-
dimensional heat equation A (x )

1-:------f----- X
0 L

For the problems considered in the previous chapter, the finite element (FE) equations
W3=l1lLLW a lx l

could be formulated directly. In general, however, the FE method is a numerical Figure 4.1 (a) Heat conduction in one-dimensional fin ; (b ) one-dimensional heat flow in
method to solve arbitrary differential equations. To achieve this objective, it is a infinitely large wall with constant thickness
characteristic feature of the FE approach that the differential equations in question
are first reformulated into an equivalent form, the so-called weak formulation. In this
chapter we shall therefore discuss the formulation in terms of so-called strong and I· dX ·I
weak forms. To facilitate this discussion we shall consider a simple differential equation,
which turns out to govern one-dimensional heat flow as well as other important H ~ C]~ H+ dH - -- x
physical phenomena like elastic bars, flexible strings, etc. To obtain a firm background,
A (x ) A (x+ dx )
it is convenient first to establish this differential equation.

a dx
Figure 4.2 Infinitely small part of fin
4.1 One-dimensional heat equation - strong form
Consider a fin with an inhomogeneous temperature distribution T(x ), as shown in
Figure 4.l(a). It is assumed that heat flows only along the fin, implying that we are fin. For this purpose, consider an infinitely small part dx of the fin, as shown in
considering a one-dimensional problem. The cross-sectional area of the fin is given Figure 4.2.
by A(x). Heat Q may be transferred to the fin across its outer surface or created In this figure, H denotes the heat inflow per unit time at position x and H + dH
internally, and Q denotes the heat input per unit time and per unit length of the fin, denotes the heat outflow per unit time at position x + dx. Both H and H + dH are
i.e. Q has the dimension [ J / s m]. Q is measured as positive if heat is supplied to the fin. considered positive when directed along the x-axis. It appears that we also have a
The heat supply Q may, for instance, be created by heating of an electric wire heat supply per unit time given by Q dx.
embodied within the fin. However, irrespective of how the heat Q is supplied, this Let us first establish a balance or conservation equation. As we only consider
heat supply can in reality only take place if a temperature variation normal to the stationary conditions, i.e. the problem is assumed to be time independent, we can
fin axis is present. That is, we do not have a one-dimensional heat problem, but we express the fact that the total heat inflow per unit time equals the total heat outflow
imagine that the thickness of the fin is small in relation to its length so that temperature per unit time. With reference to Figure 4.2, we have
variations along it are much larger than temperature variations normal to the fin H + Qdx =H +dH
axis. It is with this model approximation in mind that we consider the heat flow as
one dimensional. Another example of one-dimensional heat flow is the case of an i.e.
infinitely large wall with constant thickness as shown in Figure 4.1 (b). dH
Let us now establish the differential equation which controls the heat flow in the - =Q ( 4.1)
dx
48
so Introduction to the finite element method Strong and weak forms- 1-D heat flow Sl
By definition, we have at x =L the temperature Tis given, i.e.
H(x) = A(x)q(x)
q(x = 0) = - (k dT) =h (4.6)
where q is the so-called flux. It appears that q is the energy which passes through a dx x =O
unit area per unit time, i.e. q has the dimension [J/ m 2 s]. Moreover, the flux q is T(x = L) = g
considered positive when heat flows in the x-direction. With this expression for H, (4.7)
( 4.1) takes the form where hand g are known quantities.

B (4.2)
The so-called strong formulation of our problem is then given by the differential
equation (4.4) together with the boundary conditions (4.6) and (4.7):

Strong form of one-dimensional heat flow

We next invoke a constitutive relation, i.e. a relation which describes how heat
flows within the material. With T being the temperature and k the thermal conductivity -d ( Ak -dT) + Q = 0; 0::::;; x::::;; L (4.8)
of the material, Fourier' s law of heat conduction from 1822 then states that the flux q dx dx
is given by
q(x=O)= - (kdT) =h (4.9)
~
dx x=O
( 4.3) T(x = L) =g
~
( 4.10)

It appears that the thermal conductivity k has the dimension [J rc m s] and it is It is quite obvious that this heat conduction formulation also applies, for instance,
to diffusion problems where ions move due to differences in ion concentration. Let
convenient to take this material parameter as a positive quantity. The flux q is
measured as positive when heat flows in the x-direction. As an example consider the concentration be denoted by c, where c has the dimension [ ions/ m 3 ]. The ion
Figure 4.1 where T decreases with x, i.e. dTj dx < 0. With (4.3) this implies that the flux q is then given by the so-called Fick's law from 1855

~
flux q > 0 in accordance with the physical fact that heat flows from hotter to cooler
regions. This example illustrates why a minus sign appears in the constitutive equation
( 4.3 ). Insertion of ( 4.3) into ( 4.2) yields the following differential equation: (4.11)

-d ( Ak -dT) + Q= 0; 0::::;; X::::;; L ( 4.4) where q is now the number of ions passing through a unit area per unit time, i.e. q has
dx dx the dimension [ions/ m 2 s]. The diffusion coefficient D then takes the dimension
2
[m / s].It appears that Fick's law (4.1 1) is similar to Fourier's l.aw (4.3). Moreover,
and if the factor Ak is constant, we obtain for diffusion problems the balance equation for ions is similar to ( 4.2), i.e. both the
differential equation and the boundary conditions for diffusion problems become
analogous with those for heat conduction.
( 4.5)
Next consider electric currents for which we have the equivalent to Ohm's law
from 1827, i.e.
The one-dimensional heat equation established applies for the region considered, i.e.
0::::;; x::::;; L (cf. Figure 4.1). ~ ( 4.12)
~
In order to solve this differential equation, we require boundary conditions at the
ends of the fin. As we have a second-order differential equation we know that two
boundary conditions are required corresponding to the two ends of the fin. At the where the flux q is the electric charge passing through a unit area per unit time. We
ends we may assume either that the temperature Tis given or that the flux q is given. recall that electric charge is measured in coulombs and the ampere is defined as
If, for example, one end of the fin is completely insulated, the flux q is zero. As another coulombs per unit time. That is, the flux q has the dimension [Coulomb/ m 2 s] or
example of boundary conditions, assume that at x = 0 the flux q is given, whereas 2
[ Amp/ m ]. Moreover, in (4.12) Vis the voltage and y is the electric conductivity,
52 Introduction to the finite element method Strong and weak forms- 1-D heat flow 53
which has the dimension [Amp/Volt m], i.e. [ 1/ 0hm m]. The balance equation for The force N is given by
electric charge is similar to ( 4.2 ), i.e. the heat equation also controls electric currents.
N = A(x)a(x) ( 4.14)
where A is the cross-sectional area and a is the normal stress. With (4.14), (4.1 3)
takes the form
4.2 Axially loaded elastic bar
d
- (Aa ) + b =0 ( 4.15)
In order to illustrate that the heat equation is also encountered in other engineering dx
applications, consider an elastic bar loaded in its axial direction by body forces b
per unit axial length, i.e. b has the dimension [N/ m] (cf. Figure 4.3). The body force Let us now make use of the constitutive relation, which describes the manner in
is measured as positive if it acts in the direction of the x-axis. The cross-sectional which the material deforms due to loading. In the present case, we assume linear
area of the bar is A ( x ). elasticity , i.e. Hooke's law from 1676 states that
To establish the differential equation for this problem we again invoke a balance
principle. In the present case this balance equation takes the form of an equilibrium
condition. For this purpose consider the infinitely small part dx of the bar, as shown
I a=Ec I (4. 16)

in Figure 4.4, where N denotes the tensile force at the left end and N + dN denotes where E is Young's modulus and 6 is the normal strain, i.e. relative elongation. From
the tensile force at the right end. It appears that b dx is the total body force acting elementary m~chanics we have the kinematic relation
on the small body.
du
Equilibrium requires that the sum of all forces is equal to zero, i.e. 6=- (4.17)
dx
- N + b dx + N + dN = 0
where u is the displacement of the bar. This displacement is measured as positive
or when it is in the direction of the x-axis. Using ( 4.1 6) and ( 4.17) in ( 4.15) we obtain
dN
- + b=O (4.13)
dx -d ( AE -du) + b = 0; 0 ~ x ~L (4.18)
dx dx

Alxl and if the term AE is constant, this expression reduces to

r-----:?'L
E\---i~--- J d2 u
AE dx 2 + b = 0; 0 ~ x ~ L (4.19)
blxl
The similarity with the heat equation ( 4.8) is obvious.
~--------------------r-------__.x
In order to solve ( 4.18), boundary conditions are required. These may be given
0 L in terms of either prescribed forces or displacements at the ends. As an example,
assume that at x = 0 the force N is given, whereas at x = L the displacement u is
Figure 4.3 Axially loaded elastic bar
given, i.e.

I~ dx ·I N(x = 0) = (Aa)x =o = (AE du) =h (4.20)

dx x=O

N-~-N•dN ---x u(x = L) = g ( 4.21 )

where hand g are known quantities. The similarity of (4.20) and (4.2 1) with (4.9)
bdx
and (4.10) is obvious; we have demonstrated that the behaviour of an axially loaded
Figure 4.4 Infinitely small part of bar elastic bar is also controlled by the same equations as for one-dimensional heat flow.
64 Introduction to the finite element method Strong and weak forms - 1-D heat flow 66
4.3 Flexible string Horizontal equilibrium requires
-H + H + dH = 0
A further example, which is also similar to the equation for one-dimensional heat
flow, is the deflection of a transversely loaded flexible string. Such a string possesses i.e.
stiffness in its axial direction, but no stiffness towards bending. H =constant (4.22)
The string is shown in Figure 4.5, where w( x) is the lateral deflection in the
z-direction and p is the lateral loading per unit axial length, i.e. p has the dimension The condition for vertical equilibrium is
[Njm]. The lateral loading p is measured as positive in the z-direction. Moreover,
S denotes the tensile force in the string.
- V + p dx + V +dV = 0
In order to establish the pertinent differential equation we again make use of a i.e.
balance principle, which in the present case is obviously that of equilibrium. Consider dV
therefore an infinitely small part dx of the string, as shown in Figure 4.6. - +p=O (4.23)
At position x the tensile force in the string is S, the deflection is w and (} denotes dx
the angle that the string makes with the x-axis. The horizontal and vertical components As S is tangential to the flexible string we have
of S are given by Hand V, respectively. At position x + dx, the string force isS + dS,
H + dH (the horizontal component) and V + dV(the vertical component). Moreover, tg (}
v
= - (4.24)
it appears that p dx is the total external force acting in the z-direction on the part dx. H
Moreover

z dw dw
tg (} = - · i.e. V=H-- (4.25)
dx' dx
w(x) Inserting ( 4.25) in ( 4.23) and noting that H is constant, we get

~'---s L
X d2 w
H dxz + p = 0 (4.26)

in analogy with ( 4.8 ). Moreover H = S cos () and if we restrict ourselves to situations

LILL! f [pJ t J p(x}
where the slope () is small, we have cos(}~ 1, i.e.

H = S = constant I ( 4.27)
Figure 4.5 Laterally loaded flexible string
for small deflections. This means that the tensile force in the string is independent
of the loading. Note that this is a consequence of our assumption of small deflections.
z
With (4.27), (4.26) becomes

I· dx ·I V+dV
~S+dS (4.28)

~
H+dH
H
Referring to. Figure 4.5, the boundary conditions for the string are of the form
s ::.::=tv w w(x = 0) = 0 and w( x = L) = 0. This type of boundary condition may also occur
for heat flow. Note that no constitutive relation was involved in the derivation of
L __ _ _ _ _ _ _ _ _ _J -_ _ _ _ _ _ _ _~-----------------x

X x + dx ( 4.28 ). This is a consequence of our assumption of small deflections which implies

that S = constant, and as no change occurs in the tensile force S, the constitutive
Figure 4.6 Infinitely small part of string relation is not involved.
 Strong and weak forms- 1-D heat flow 57
56 Introduction to the finite element method
Use of this expression in (4.29) implies that
4.4 Weak form of one-dimensional heat flow
Having established the differential equation and boundary conditions (i.e. the so-called L dv Ak dT dx = [vAk dTJL + fL vQ dx (4.34)
f 0 dx dx dx 0 0
strong form) of one-dimensional heat flow (cf. (4.8)- (4.10)), and having illustrated
that this formulation also covers other physical phenomena, we shall now introduce We claimed previously that the weight function v(x) is arbitrary. In fact, this is
the important concept of a weak formulation. The weak formulation is in fact a not completely true, since we must require that the derivations above are meaningful.
reformulation of the strong form and it is from this weak form that the FE approach We shall not probe into this issue, but just note that the mathematical restrictions
is established. that we must place on the function v(x) are clearly quite modest. We refer to Becker
To establish the weak form of the strong form given by ( 4.8)- ( 4.10), multiply et al. (1981), Hughes (1987) and Strang and Fix (1973) for a detailed mathematical
( 4.8) by an arbitrary function - the so-called weight function - v(x) to obtain treatment.
It is important that, due to the integration by parts, ( 4.34) contains terms related
v[ddx ( Ak ~:) + QJ= 0 to the boundary, i.e. the ends of the fin. These terms are given by the first expression
on the right-hand side of (4.34). Now, using q = -k dT/ dx as well as the boundary
We may even integrate this equation over the pertinent region, i.e. 0 :$; x :$; L, to obtain condition ( 4.9) we get

f: v[ :x ( Ak ~:) + QJdx = 0 (4.29) dTJL = ( vAk-

[ vAk-
dx o
dT)
dx x=L
- ( vAk -dT)
dx x= O
= - (vAq)x=I. + (vA)x=oh
( 4.35)
In order to evaluate this expression further, we recall from elementary analysis
the rule for integration by parts. By definition we have where the flux q(x = L) is unknown, whereas the flux q(x = 0) = h is known.
b-dy dx = fb dy = [y(x)Jab (4.30)
Using (4.35) in (4.34) we obtain the weak form of one-dimensional heat flow:
fa dx a

Put Weak form of one-dimensional heat flow

y(x) = cf>(x)t{!(x) Ldv dT fL
- Ak - dx = -(vAq)x =L + (vA)x= oh + vQdx (4.36)
f 0 dx dx 0
i.e.
T( x = L) = g (4.37)
dy =de/> t/1 + c/> dt/1 ( 4.31)
dx dx dx
where (4.37) is just a repetition of ( 4.10) and where the weight function
With (4.31), (4.30) implies that v is an arbitrary function.
The weak form was obtained from the strong form by suitable manipulations,

i.e.
fa
b (de/> t{! + cf> dt{!) dx = [cf>(x)l{!(x)]:
dx dx
i.e. the strong form implies the weak form. It seems natural to investigate whether
the converse statement is true, i.e. whether the weak form implies the strong form.
The answer is in fact affirmative, as will be shown next. In order to prove that the
weak form implies the strong form, the key point is that ( 4.36) is required to hold
not only for one function v(x) but also for any choice of v(x).
fabc/> dt{!
dx
dx = [ c/>t/1 J: - fb de/> t/1 dx
a dx
(4.32) To show that the weak form implies the strong form, we accept ( 4.36) and ( 4.37).
We can eliminate the term on the left-hand side of( 4.36) by using the rule of integration
by parts given in ( 4.33 ). Expression ( 4.36) then takes the form
With this result we conclude that

L v-d ( Ak -d T) dx = [ vAk-
d TJL - fL -
dvAk d- T dx (4.33)
dTJL - fL v - Ak-
[ vAk -dx dx dx
d( dT) dx =-(vAq)x~L+(vA)x=oh + fL vQdx
f 0 dx dx dx 0 0 dx dx 0 0 0
 Strong and weak forms - 1-D h ea t flow 59
58 Introduction to the finite element method
As the flux may be prescribed at the boundary, it is therefore called a natural boundary
I.e.
condition. Occasionally, the term Neumann boundary condition is used. On the other

f: v[:x(Ak~:) + QJdx
hand, the boundary condition given by ( 4.10) and ( 4.37), which prescribes the value
of the variable itself, is called an essential boundary condition. Occasionally the term
Dirichlet boundary condition is used. Boundary conditions like (4.9) and ( 4.10), which
- (vA)x=L(k dT
dx
+ q)x=L + (vA)x =o(k dT
dx
+h)
x=O
=0 ( 4.38) contain both natural and essential conditions, are also called mixed boundary
conditions.
We recall that the weight function v is an arbitrary function. We can therefore put We emphasize that ( 4.36) of the weak formulation is identical with the fulfilment
of both ( 4.8) and ( 4.9) and vice versa: whereas ( 4.8) expresses the balance principle

"~¢[d~(Ak!:)+Ql where ¢{:~: ~:~<L

within the body, ( 4.9) expresses the balance principle at the boundary.
(4.39)

Thisimpliesthatv(x = 0) = v(x = L) = 0. Using(4.39)in(4.38)wethereforeobtain 4.4.1 Advantages of the weak formulation compared with the

f: ~{ ddx ( Ak ~:) + Q J dx = 0
strong form

We have proved that the weak and strong formulations are identical, so one may
ask why we have taken the trouble to obtain the weak form. The answer is fundamental
from which we conclude that
for the establishment of the FE method: it is on the weak form that the FE formulation
~(Ak dT) + Q = 0; 0s X s L (4.40) is based.
In order to understand this point consider the differential equation in the strong
dx dx
form, i.e. ( 4.8 ). It appears that the unknown function Tis differentiated twice. The FE
We immediately observe that this is the differential equation in the strong form (cf. approach is an appr oximate method, so in one way or another we have to replace
(4.8)). Moreover, using (4.40) in (4.38) we are left with the true T-function by an approximate one. If this approximation is made directly

- (vA)x =L(k dT
dx
+ q)x=L + (vA)x =o(k dT
dx
+h)
x= O
=0 (4.41)
in ( 4.8 ), we need to deal with an approximating function, which is at least twice
differentiable. In the weak form, however, only the first derivative of the temperature
T enters ( cf. ( 4.36)). That is, if we choose the weak form as the basis for the
As the weight function v is arbitrary, we now choose a v-functio n for which approximation, we may deal with approximating functions which only need to be
v(x = 0) = 0 and v(x = L ) :f. 0. From (4.41 ) we then conclude that differentiable once. This aspect clearly favours the weak form compared with the

-(k dT)
d x x=L
= q(x = L)
strong form, and it also suggests the terminology of weak and stro ng forms.
Another point, closely related to the matter discussed above, is that the weak
form provides in fact a more general formulation than the strong form. This statement
Likewise, choosing a v-function for which v(x = 0) :f. 0 and v(x = L) = 0, we obtain may seem contradictor y, as we have just proved that the two forms are equivalent.
from ( 4.41 ) that However, in doing so we implicitly assumed that the variable T could be differentiated
as many times as was necessary, i.e. twice. What is of importance is that the weak
- (kdT) =h formulation ( 4.36) and ( 4.37) holds in an unchanged form even when discontinuities
dx x=O occur, whereas the presence of discontinuities requires a modification of the strong
form.
which is exactly the boundary condition given by ( 4.9).
Therefore, we have shown that the weak form given by ( 4.36) and ( 4.37) implies To illustrate this issue, consider again the fin of Figure 4.1, but assume now that
the strong form given by ( 4.8)-( 4.10). Moreover, we already know that the strong the thermal conductivity k experiences a discontinuity at point x = a, as shown in
form implies the weak form, i.e. the weak and strong formula tions of our problem Figure 4.7, where k = k 1 holds in region 1 to the left of point x = a and k = k 2 holds
in region 2 to the right of point x = a.
are identical.
Let us return to the boundary conditions (4.9) and (4.10). We have seen that Despite the discontinuity of the thermal conductivity at point x = a, the amount
the flux q (or h) emerges by itself in the boundary terms of the weak formulatio n. of heat leaving region 1 at point x = a must equal the amount of heat entering region
60 Introduction to th e finite element method Strong and weak forms- 1-D heat flow 61
Region 1 Region 2 For this purpose, multiply (4.43 ) by an arbitrary function v - the weight
tmf k = k ~0&r::,:s,~Y:2~~m
1 . X
function - and integrate over the pertinent region, i.e.
x=a
( 4.47 )
Figure 4.7 Discontinuity of thermal conductivity
Likewise for region 2 we obtain from ( 4.44)
T
( 4.48)

Integrating (4.47) by parts similarly to ( 4.33) we derive

I a dv dT [ dT]a fa
I
I
- Ak 1 - d x = vAk 1 - + vQ d x ( 4.49)
I f 0 dx dx dx 0 0
Reg ion 1 I Reg io n 2
f* k= k1 ~?>q:>\~~~'-.9a~
·:};:'t----· X
In a like manner, we obtain from (4.48)

L dv Akz dT dx = [vAkz dTJL + IL vQ dx (4.50)

L
I a dx dx dx a a
Figure 4.8 Non-smooth temperature distribution at x = a
In (4.49) and (4.50), the derivative dTi dx appears in the integrands. As Tis non-
smooth at point x = a ( cf. Figure 4.8) the derivative d T I d x experiences a discontinuity
2 at point x = a, i.e. at this point. We then obtain the distribution of d T I d x as shown in Figure 4.9.
The important point is that even though d T l d x experiences a discontinuity, an
- Ak 1 -dT) = - ( Ak 2 -dT) (4.42) integration of dTi dx over the region is still well defined. With this observation in
( dx x=a dx x=a mind, one may add (4.49) and (4.50) to obtain
As k 1 -# k 2 the derivative d T / dx takes different values to the left and right of point
x = a. As the temperature T must be a continuous function, we obtain the temperature L dv Ak dT dx = ( vAk 2 dT) - ( vAk 1 dT ) + fL vQ dx ( 4.51)
f o dx dx dx x= L dx x=O o
distribution shown in principle in Figure 4.8. It appears that the temperature T is
continuous, but non-smooth at point x = a. where use is made of (4.42) and where k = k 1 fo r 0 sxs a and k = k 2 fo r a< x s L.
It follows directly that in regions 1 and 2, the differential equation ( 4.8) still
holds, i.e.
dT/dx
~(Ak
dx
1 dT)
dx
+ Q = 0; Osxsa ( 4.43)

~(Ak2 dT) + Q = 0; a sxsL ( 4.44)

dx dx
L
Moreover, we again have the boundary conditions ( 4.9) and ( 4.10), i.e.

q(x = 0) = - (k dT)1 = h ( 4.45) I

v
I
dx x=O
T(x = L) = g ( 4.46)
Equations ( 4.42)- ( 4.46) now constitute the strong form of our problem. Next let us
establish the weak formulation. Figure 4.9 Discontinuity of dT / dx at x =a
62 Introduction to the finite element m eth od Strong and weak forms- 1-D heat flow 63

Using the boundary condition (4.45) as well as q = - k 2 dTj dx, we get Table 4.1 Examples of second-order differential equations
Differential equation Physical problem Quantities Constitutive law
L dv Ak dT dx = - (vAq)x =L + (vA)x =oh + IL vQ dx ( 4.52)
I 0 dx dx 0 -d ( Ak -dT) + Q =O One-dimensional T = temperature Fourier
dx dx heat flow A = area q = - kdT/ dx
Moreover, we also have the essential boundary condition ( 4.46), namely
k = thermal conductivity q = heat flux
T(x = L) = g ( 4.53) Q = heat supply

We observe that the weak formulation (4.52) and (4.53) is exactly the same as the -d ( AE -du) + b = O Axially loaded u = displacement Hooke
dx dx elastic bar A= area u = E du fdx
one given by ( 4.36) and ( 4.37) even though we are now considering a discontinuous
problem. By comparison, the strong forms of the continuous problem ( 4.8)- ( 4.10) E = Young's modulus u = stress
b = axial loading
and of the discontinuous problem ( 4.42)- (4.46) differ.
In this sense, we have shown that the weak form is more general than the strong d2w
S- + p= O Transversely loaded w = deflection
form, as the weak form applies irrespective of possible discontinuities. For a more dx 2 flexible string S = string force
detailed discussion of these matters the reader may consult Becker et al. (1981), and p = lateral loading
for a general in-depth discussion of the weak formulation reference may be made to
Zienkiewicz and Taylor (1989). -d ( AD -de) + Q = O One-dimensional c = iron concentration Fick
dx dx diffusion A = area q = - Ddc fdx
D = diffusion coefficient q = ion flux
Q = ion supply

-d ( AydV)
- +Q = O One-dimensional V = voltage Ohm
4.4.2 Alternative derivation of heat equation dx dx electric current A= area q =-y dV f dx
y = electric conductivity q = electric charge flux
Previously we established the differential equation for one-dimensional heat flow by Q = electric charge supply
considering an infinitely small part dx of the body ( cf. Figure 4.2). For later purposes 2
D -
-d ( A - dp) + Q = O Laminar flow in pipe p = pressure q = - (D 2 f 32Jl) dp f dx
it is of interest to establish this differential equation directly by using simple rules of dx 32)l dx ( Poiseuille flow) A = area q = volume flux
integration. D = diameter q = mean velocity
The fin is shown again in Figure 4.10, where for simplicity we disregard any Jl = viscosity
possible discontinuities. Let us consider the arbitrary region a < x < b. Q = fluid supply
The balance or conservation principle for this region states that the heat per unit

(Aq)x=a + r
time entering the region equals the heat per unit time leaving the body, i.e.

Q dx = (Aq)x=b

which can be written as

By definition we have

[
Ak dT]b = fb ~(Ak dT) dx
dx a a dx dx
and use of this expression in (4.54) implies that
fQdx =[AqJ: or fQdx=- [Ak~:J: ( 4.54)

As this integral holds for an arbitrary region, i.e. arbitrary a- and b-values, we conclude
that
~=t:E~]-·x d ( Ak dT)
dx dx + Q = 0; 0 ~ x ~ L
0 0 b
L

Figure 4.10 Fin with arbitrary region a < x < b which is exactly the differential equation established previously in ( 4.8 ).
64 Introduction to the finite element method

In the present one-dimensional case this alternative derivation does not provide
any significant advantages over the approach adopted in Figure 4.2. For two- and
5
three-dimensional problems, however, we shall see later that an analogous formulation
to that illustrated above results in great simplifications when establishing differential
Gradient, Gauss' divergence
equations for various physical problems.
theorem and the Green-
Gauss theorem
4.5 Concluding remarks
We have seen that various physical problems lead to the same type of differential
equation and boundary conditions and some examples are given in Table 4.1. The
differential equation and boundary conditions together were referred to as the strong
form of the problem. We showed that this strong form could be reformulated into
an equivalent weak form.
It was emphasized that the weak form is the one on which the FE approach is For a one-dimensional situation, we showed in Chapter 4 that in order to obtain
based. The reason is that the order of differentiation of the unknown is lower in the the weak form from the strong form, integration by parts was necessary (cf. (4.33 )).
weak form than in the strong form, thus facilitating the approximation process. In two or three dimensions, a similar type of integration needs to be performed, and
Moreover, the weak form applies without changes to continuous as well as for this purpose we shall make use of the Green-Gauss theorem, which is based on
discontinuous problems, implying the important conclusion that the FE method also Gauss' divergence theorem. These theorems are of fundamental importance for many
holds for continuous as well as discontinuous problems. applications. Before presenting a heuristic proof for Gauss' divergence theorem and
We discussed these aspects and their various ramifications for a specific differential the closely related Green - Gauss theorem, we shall discuss the concept of the gradient
equation since the same type of arguments can be put forward for more complicated of a function.
situations like those encountered in two- and three-dimensional problems. Therefore,
the conclusions of the present chapter will carry over to the more advanced problems
that will be dealt with in later chapters.

5.1 Gradient
Consider the two-dimensional quantity </> which depends on the coordinates x
andy, i.e. </> = <J>(x, y). As</> varies with position and as</> only comprises one quantity,
we say that </> constitutes a scalar field over the region in which x and y are defined.
As an example, </> may be taken to be the temperature distribution over a
two-dimensional body and Figure 5.1 shows a principle sketch of the </>-distribution.
To investigate how </> = <J>(x, y) varies with position, we use the chain rule to
obtain
o<J> o<J>
d</>=-dx + - dy (5.1)
ax ay
What ( 5.1) expresses is illustratea m Figure 5.2, where point P 1 has coordinates ( x, y)
and at that point we have </J(x, y). At the adjacent point P 2 with coordinates
(x + dx, y + dy) the function takes the value </> + d</>, where d¢ is given by (5.1).

65
66 Introduction to the finite element method G r adient and Gauss theorems 67
y

vq> /...._e /
. •1/dx
I
l

llJ----=t-
i"(J, 1I !
,_
L---------------------- x
Figure 5.3 Angle{) between vectors V 4> and dx

-~~, y

Figure 5.1 Distribution of 4> over region in xy-plane

P ( x+dX , y+dy )
2
• 4>+d4>

.,,,,_/ldy
4> dx
L------------------------------------x
Figure 5.4 Contour curves in the xy-plane; C 1 , C 2 and C 3 are constants

L-------------------------------x ( 5.2 ), ( 5.1) can be rewritten according to

Figure 5.2 Change of 4> with position

de/>= [ acj> acj>J [dx] = (Vr/J )T dx (5.3)
ax ay dy
i.e. d¢ is the scalar product of VcJ> and dx. We recall from elementary vector algebra
•
Let us introduce column matrices Vc/> and dx defined by that this scalar product is given by
de/>= (Vc/>)T dx = IV¢11dxl cos 0 (5.4)
where IV¢ 1 and ldxl are the lengths of the corresponding vectors (cf. (2.18)) and
(5.2) 0 is the angle between the two vectors; see Figure 5.3.
As cJ> = ¢(x, y), it is obvious that we can identify so-called contour curves along
which cJ> remains constant. This is achieved by putting ¢ (x, y) = C, where C is a
constant, and it follows that this expression provides a relation between x and y, i.e.
it describes a curve - a contour curve - in the xy-plane. Such contour curves are
From Figure 5.2 it appears that the column matrix dx is in fact a vector with illustrated in Figure 5.4.
components dx and dy in the x- and y-directions, respectively. Likewise, we can Let us now assume that we move along a contour curve for which ¢ is constant.
consider the column matrix Vel> as a vector with components acj> j ax and acj> jay in For such a change in position the vector dx is directed tangentially to the contour
the x- andy-directions, respectively. The vector Vc/> is called the gradient of cj>. With curve ( cf. Figure 5.5 ).
68 Introduction to the bnite element method Gradient and Gauss theorems 69
y Let us now assume that we are at the position given by point P 1 of Figure 5.6, i.e.
the gradient V'¢ is given and fixed. For any given length ldx l of dx we conclude from
( 5.4) that the largest possible increase of ¢ occurs when d x is in the direction of the
gradient V'¢.
The change of ¢ with position is given by (5.3). However, fo r later purposes it
is convenient to rewrite this expression slightly. The change of position is given by
the vector dx. Define the length of dx by dm, i.e.

(5.6)

It is then possible to define a unit vector m in the direction of dx. W ith mX and my
L---------------------------------- x
being the components of the m-vector, this vector is given by
Figure 5.5 Change of position along contour curve
m = [mx] = -dx
my dm
and Im l = 1 ( 5.7)

Dividing (5.3 ) by dm and using ( 5.7 ), it follows that

(5.8)

We o bserve tha t d¢ / d m is the slope of ¢ in the direction of the unit vector m. This
slope is illustrated in Figure 5.7.
Figure 5.6 Two infinitely close contour curves
The results above can be generalized directly to three dimensions where
¢ = cp(x, y, z). From (5.3 ) we have
Moreover, as ¢is constant along the contour curve we have from (5.4) that
d¢ = (V'¢)T dx (5.9)
d¢ = (V' ¢)T dx = IV'¢ IIdx I cos e = 0
which can be fulfilled only if cos e = 0, i.e. the two vectors are orthogonal. As dx is
tangential to the contour curve, we conclude that the gradient V'¢ is orthogonal to
the contour curve. However, given a certain contour curve this leaves us with two •
possible and opposite directions of the vector V'¢.
To determine the correct direction of these two possibilities, consider Figure 5.6
where two infinitely close contour curves are shown and where C 1 < C 2 •
Assume that we are at point P 1 and move to the adjacent point P 2 as indicated
by the vector dx. Since¢ is larger at point P 2 than at point P 1 , we have from ( 5.4) that
d¢ = (V'¢)T dx = IV'¢ 11 dx l cose> 0 (5.5)
This expression can only be true if cos e > 0, i.e. the gradient V'¢ is directed as shown
y
in Figure 5.6. We conclude the following :

The gradient V' ¢ is orthogonal to the contour curve and the direction
of V'¢ is in the direction of increasing ¢-values.
Figure 5.7 Illustration of slope df/;/ dm
70 Introduction to the finite elemen t method Gradient and Gauss theorems 71
where the gradient V<P and dx are given by y

0
r- o<P. y= d
dx
OX
o</J
V</J = oy
dx = dy ( 5.10) X= X (
1
y)

o</J
dz
.. az y= c
c
From (5.9) we conclude that L---------------------------------------x
Figure 5.8 Integration over region A

( 5.11)
By definition we have

where the unit vector m is given by

f a<jJ dA = ff o</J dx dy = fd ( Jxz(y) a<jJ dx) dy

m ~ [ : } : : and dm ~ ldxl; ie lm l ~I (5.12)

A ax

= rA ax

[</J(x,
c

y)]~:::~:gl dy = r
x,(y) ax

[</J(x2(y), y) - </J(xl(y), y)] dy

= r </J(x2(y), y) dy + f </J(xl (y), y) dy

It appears that the first term on the right-hand side is the integral taken along the
(5.13)

boundary x = x 2 (y) in the counter-clockwise direction ( cf. F igure 5.8), whereas the
5.2 Gauss' divergence theorem and the Green-Gauss second term on the right-hand side is the integral taken along the boundary x = x 1 (y)
theorem in the counter-clockwise direction. By definition, it then follows that ( 5.13) reduces to

The fundamental divergence theorem of Gauss and the closely related Green- Gauss
theorem will now be derived in a heuristic manner. More mathematical proofs may
be found in Kaplan (1981), Kreysig (1979) and Sokolnikoff and Redheffer (1958).
fA
o</J dA
ax
= J. </J(x, y) dy
:r~
where the boundary integral is taken in the counter-clockwise direction along the
( 5.14)

For the function <P = </J(x, y) defined over the area - i.e. region - A, consider the boundary !t'. In exactly the same manner, it follows that for a function 1/1 = 1/J(x, y)
area integral we find

f A
a<P dA
OX fA
aljl dA
ay
=- J.
:r~
1/J(x, y) dx

where the boundary integral is again taken in the counter-clockwise direction.

( 5.15)

The region is shown in Figure 5.8. The maximum and minimum y-values of the Let us rewrite expressions ( 5.14) and ( 5.15) by introducing the incremental arc
region are indicated by points D and C, where the y-values are given by y = d and length d!t' along the bo undary. Recalling that we perform the boundary integration
y = c, respectively. These two points divide the boundary into two curves : one in the counter-clockwise direction, we have the situation sketched in Figure 5.9.
described by x = x 1 (y) and the other by x = x 2 (y) (see Figure 5.8). In this figure, n denotes the unit vector normal to the boundary and directed
72 Introduction to the finite element method Gradient and Gauss theorems 73
y

X
Figure 5.10 Region A with boundary 2' and outer unit vector o
Figure 5.9 Outer unit normal vector o and incremental arc length d.P
Using (5.19) in (5.14) and (5.15) we get (see Figure 5.10)

out of the region. This vector is given by

f A
o</J dA =
OX
1 </Jnx d£'
J9' (5.20)

n= [::l lnl = (n; + n;) 112 = 1 ( 5.16)

f A
ot/1 dA = 1
oy 'J9'
t/lny d£' ( 5.21)

where nX and ny are the components of the n-vector along the x- and y-axes, In (5.14) and (5.15) the boundary integral should be taken in the counter-clockwise
respectively. By definition, d£' is a positive quantity given by direction. In (5.20) and ( 5.21 ), however, as d£', irrespective of the integration direction,
is a positive quantity and as the terms </Jnx and t/fny are independent of the direction
1 d£'=(dx 2 +dy 2)1' 2 1 (5.17) of integration, the direction of the boundary integration is arbitrary.
Now let us define the arbitrary vector q by
where dx and dy indicate the change of position along the boundary, i.e. d£' is the
length of the increment along the boundary (cf. Figure 5.9). This expression ford£'
(5.22)
may also be written as

(~~YJ'
2
dx)2]1/2 where qx and qY are the components in the x- and y-directions, respectively. Adding
d£' = ldx{ 1 + > 0 or d£' = Idyl [ 1 + ( dy > o (5.20) and (5.21) and using (5.22), we obtain Gauss' divergence theorem
( 5.18)

if dx -1= 0 or dy i= 0, respectively.
From Figure 5.9, the angle 0 can be identified in the triangles CDE and FGE.
L div q dA = f9' qrn d£' ( 5.23)

That is, for the length of the sides of these triangles we have where div q is the divergence of q defined by
CE FE . CD FG
cosO= - = -; smO= - = - . oqx oq
ED EG ED EG dtvq = - + _Y ( 5.24)
ax ay
From Figure 5.9 these lengths (that are positive quantities) can be identified as
dx ny dy nx The one-dimensional counterpart of ( 5.23) is given by (4.30 ).
- d£' = 1; d£' = 1 Assume now that instead of q we consider the vector </Jq where <P is a scalar.
Using the definition (5.24) we find that
where CE = ldxl = - dx since dx is negative in the present case. We then conclude that
a a oqx oqy o</J o</J
dx = - nyd£'; dy = nx d£' (5.19) div( </Jq) = OX ( </Jqx) + oy ( </Jqy ) = <P OX + <P oy + OX qx + oy qy
74 Introduction to the finite element method Gradient and Gauss theorems 75
which, using the definition of the gradient V¢ (cf. (5.2)), can be written as z

div(</Jq) = ¢ div q + (V</J)Tq (5.25) s

Inserting </Jq instead of q in ( 5.23) and making use of ( 5.25 ), the Green - Gauss theorem
is obtained according to

(5.26) X

Figure 5.11 Volume V with surface boundary Sand outer unit vector n

When integrating by parts in the one-dimensional case, the form given by ( 4.32) is
Let the unit vector n normal to the boundary and directed outwards be given by

:J
obtained. The Green- Gauss theorem can be seen as the two-dimensional counterpart

~ ~ ~
for integration by parts, and just as for the one-dimensional case, it appears from
(5.26) that boundary terms emerge as a result of this integration by parts.
Sometimes, it is important to write the Green- Gauss theorem in a slightly n [ InI (n; + n; + n;)'" I (5.31)
different way. If ! he arbitrary vector q is chosen so that qx = tjJ and qY = 0, (5.26)
reduces to where nx, ny, n, are components in the x-, y-, and z-directions, respectively. Gauss '
divergence theorem then reads
( 5.27)
( 5.32)
Alternatively, if q is chosen such that qx = 0 and qY = t/J, (5.26) implies that
where the divergence of q is integrated over the volume V and where the quantity
qT n is integrated over the surface, i.e. the boundary, of the region V (see Figure 5.11).
Finally, the Green-Gauss theorem takes the form

( 5.33 )
The fundamental results of (5.23) and (5.26) carry over to three dimensions; see
for instance Kaplan (1981), Kreyszig (1979) and Sokolnikoff and Redheffer (1958).
That is, if the arbitrary vector q is given by An alternative form of this theorem is obtained by first choosing the arbitrary
vector q as qT=[t/1,0,0], then choosing qT=[O,tjJ,O] and finally putting
qT = [0, 0, tjl]. The result is
(5.29)

where qx, qY and q, are the components in the x-, y- and z-directions, respectively,
then the divergence of q becomes ( 5.34)

div q = oqx + oqy + oq. ( 5.30)

ox oy az
 Strong and weak forms- 2-D and 3-D heat flow 77
6 y

Strong and weak forms q n

two- and three-dimensional

heat flow ~
~---------------------- x

Figure 6.1 Flux q. at boundary

by the unit vector n normal to the boundary and directed outwards, i.e.
In Chapter 4 we derived the one-dimensional heat equation and discussed its strong
and weak forms. Moreover, it was demonstrated that various important physical
phenomena are controlled by formulations analogous to the heat flow problem. It
was emphasized that the FE method is based on the weak formulation. In the present ( 6.2)
chapter, a similar discussion will be carried out for the heat flow equation in two
and three dimensions. It will turn out in this discussion that extensive use is made
of Gauss' divergence theorem as well as of the Green-Gauss theorem. It is of interest to determine the total amount of heat which passes through a unit
area of the boundary per unit time. This amount is denoted by q. and it is called
the flux. It is essential to distinguish the flux q. from the flux vector q, but clearly
the two quantities are related. The flux vector q, which indicates the direction of the
heat flow, is also illustrated in Figure 6.1. It is obvious that the component of q
6.1 Heat flux vector-constitutive relation tangential to the boundary does not contribute to the flux q. which passes through
the boundary. Rather, the flux q. is given by the component of q in the direction of
In one dimension the amount of heat passing through a unit area per unit time was the unit vector n, i.e. the flux q. is given by
given by the flux, which was related to the derivative of the temperature according
to the constitutive relation provided by Fourier's law (cf. ( 4.3 )). (6.3)
For two- or three-dimensional problems heat may flow in various directions and
this flow may conveniently be described by the heat flux vector q. This flux vector q
has the direction of the heat flow and its length expresses the heat per unit time From Figure 6.1 it follows that q. is positive when heat leaves the boundary at the
which passes through a unit surface area perpendicular to the direction of heat flow. specific position considered. We are now in a position to evaluate the components
We have qx, qY and qz of q (cf. (6.1 )). Considering, for instance, the flux q. in the x-direction,
we have nT = (1, 0, 0), i.e. (6.3) gives q. = qx. Similar interpretations apply for the
components qY and qz.
(6.1) It is obvious that the heat flux vector q is related to the temperature gradient
VT and this relation is expressed by the constitutive equation. F or this purpose the
generalization of Fourier's law for one-dimensional heat flow, (4.3), to two and three
in two and three dimensions, respectively, where qx , qY and qz are the components dimensions becomes
of q in the x-, y- and z-directions, respectively. These components have the dimension
[J/ m 2 s].
q = - DVT (6.4)
Referring to Figure 6.1, we now consider the boundary which is characterized

76
78 Introduction to the finite element method Strong and weak form s - 2-D and 3-D hea t flow 79

The temperature gradient VT is given by (5.2) and (5.10), i.e. for all VT # 0, which proves that the constitutive matrix D is positive definite ( cf.
(2.66)). From (2.67) we then conclude that D is non-singular, i.e.
aT
ax I I (6.9)

vr~ [~:l
aT · vr~
det D # 0
aT (6.5) In the most general form D may be non-symmetric, but for convenience and because
ay it works well in practice, we assume that D is symmetric, i.e.
ay aT
az ( 6.10)

in two and three dimensions, respectively. Moreover, in (6.4) D is the so-called Let us investigate the constitutive matrix D in more detail. Suppose, for instance,
constitutive matrix which contains information about the ease with which heat flows that in the two-dimensional case the temperature only changes in the x-direction.
in different directions. That is, D is given by The temperature gradient is then given by

(6.6)

With ( 6.4) and ( 6.6) we then obtain

in two and three dimensions, respectively. In Fourier's law for one-dimensional heat
aTjax.
[::J [~:
flow, (4.3), the flux depends linearly on It appears that Fourier's generalized
law, (6.4), presents the most general form for which the flux vector q depends linearly
on the derivatives of the temperature T. If the constitutive matrix D depends on q= = .. :;]
position, i.e. D = D(x, y, z) then the material is referred to as inhomogeneous;
otherwise we speak of a homogeneous material.
yx ax
The implication of the constitutive relation ( 6.4) is illustrated in Figure 6.2. The It appears that even though there is no temperature change in the y-direction we
temperature gradient VT is directed towards hotter regions and physical arguments still have a heat flow, qY # 0, in that direction. In general, this type of coupling effect
imply that heat flows from hotter to cooler regions. Consequently, the scalar product is caused by the off-diagonal components of the constitutive matrix D. It appears
of the vectors VT and q is negative, i.e. that D as given by (6.6) applies to very general, so-called anisotropic materials.
A somewhat simpler form emerges if D is taken as
(VT)Tq < 0 (6.7)
Using relation ( 6.4) in this expression, it follows that
D = [kxx O ]· D =
kxx
0
0 0]
k 0 (6.11)
0 k ' [
yy
(6.8) .
yy
0 0 k,,
where the thermal conductivities kxx• kYY' kzz in the x-, y- and z-directions may be
y
different. The forms of D given by ( 6.11) correspond to an orthotropic material and
Hot we obtain from ( 6.4) and ( 6.11) that

V'T -k aT ax
q----_/ -kXX aT]
ax -k aT
XX

aT ; ( 6.12)
Cool
q= q = ayyy
[ - k -ay
- k aT
yy
L------------------------X
zz az
Figure 6.2 Flux vector q and temperature gradient VT
80 Introduction to the finite element m ethod Strong and weak forms - 2-D and 3-D heat flow 81
y
6.2 Heat equation for two and three dimensions- strong
Hoi form
Let us now derive the differential equation which controls heat flow in more
dimensions. As in the discussion of the one-dimensional heat equation, we may
consider an infinitely small part of the body similar to Figure 4.2. However, in the
present case it is much simpler and more elegant to use a derivation analogous to
Cool
the alternative derivation given in Chapter 4 for one-dimensional heat flow.
L--------------------------x The key point here is the fundamental divergence theorem of Gauss. Let us
Figure 6.3 Flux vector q and temperature gradient VT for isotropic materials consider a two-dimensional body and let us assume that Q is the amount of heat
supplied to the body per unit volume and per unit time; that is, Q is positive when
heat is supplied to the body and has the dimension [Jj m 3 s]. As we only consider
stationary (i.e. time-independent) problems, the balance or conservation principle
states that the amount of heat supplied to the body per unit time equals the amount
For an isotropic material, in which the thermal conductivity is the same in all of heat leaving the body per unit time. We then conclude that
directions, (6.11) reduces to

(6.13)
t Qt dA = f.SI' qnt d.Sf (6.19)

where t = t(x, y) denotes the thickness in the z-direction of the body located in the
xy-plane; see Figure 6.4.
From (6.3) we have that q" = qTn, i.e. Gauss' divergence theorem (5.23) yields
which can be written as
D = kl (6.14) J. tqn d2' = J. tqTn d2' = J. (tq)Tn d.P =I div(tq)dA (6.20)
i.e. we have only one material parameter, the thermal conductivity k. From (6.8) we
J.se J.se J.se A

conclude that k is a positive quantity. In practice, the most common situation is the With (6.20), (6.19) takes the form
isotropic one given by (6.14). With (6.14), (6.4) reduces to
q = -kVT (6.15)
L [tQ - div(tq)] dA = 0

i.e. the vectors q and VT have opposite directions as illustrated in Figure 6.3. In this expression, we did not specify the region A. In fact, this region may be taken
With (6.4) we get from (6.3) as any arbitrary part of the region for the entire body. Therefore, as the region A is
arbitrary, we conclude that
q" = - (VT)TDn (6.16)
utilizing the fact that D is symmetric (cf. (6.10)). If Dis given by (6.11), we obtain I div(tq) = tQ (6.21)
from (6.2) and (6.16)
which is the balance principle for the two-dimensional body. If the thickness t is
qn = - kxx -
oT nx - oT
kyy- ny ( 6.17)
ox oy
for two dimensions and
__ll
qn = -
or
kxx - nx -
oT
kyy - ny -
oT
kzz - nz (6.18) T
ox oy oz
for three dimensions. Figure 6.4 Two-dimensional body with thickness t located in the xy-plane
82 Introduction to the finite element method Strong and weak forms - 2-D and 3-D heat flow 83
constant, we get y T= g

divq = Q (6.22)

where, according to ( 5.24 ), the divergence of q is given by

q n= h
. oqx oqy
dIvq= - + - ( 6.23)
ox ay
Using the constitutive relation ( 6.4) in the balance equation ( 6.21 ), it follows that
div(tDVT) + tQ = 0 in region A (6.24) Figure 6.5 Two-dimensional region A with boundary .!£ = !£h + !t'g
where region A is the entire region of the body. If the constitutive equation D is
given by ( 6.11 ), ( 6.24) reduces to We note that it is not possible to prescribe the flux qn and the temperature Tat
the same position.
axa ( tk xx -ar)
-
ax + -oya ( tk yy -ar)
oy + tQ = 0 (6.25) Whereas the boundary conditions (6.29) and (6.30) are typical for applications,
it may also happen that the boundary condition is prescribed in terms of a linear
For reasons that will be revealed in a moment, this equation Is called the combination of the flux qn and the temperature T, i.e.
quasi-harmonic equation. For isotropic materials, (6.13), we obtain ( 6.31 )

~ (tk ar) + ~ (tk ar) + tQ = o (6.26) where a and {3 are known quantities. A boundary condition in the fo rm of (6.31 ) is
ax ax oy oy typical when convection occurs along the boundary; see, for instance, Carslaw and
Jaeger ( 1959). Here, we shall mainly be concerned with boundary conditions in the
Furthermore, if the product tk is constant, (6.26) reduces to
form given by (6.29) and (6.30).
oT + -
-
2
o T+ -Q =0
2
( 6.27)
With (6.24), (6.29) and (6.30) we have obtained the strong form of the heat
ox oy k 2 2 problem:

This expression is the well-known Poisson equation and we shall show later in Chapter
14 that this differential equation also applies for the torsion of non-circular elastic Strong form of two-dimensional heat flow
shafts. If the internal heat supply is zero, i.e. Q = 0, (6.27) becomes
div(tD VT) + tQ = 0 in region A (6.32)
o T +o-T= 0
2 2
- (6.28) on ;t;. (6.33)
ox oy 2 2

T=g on 2g (6.34)
which constitutes the well-known Laplace equation. Solutions to the Laplace equation
are termed harmonic functions and this suggests why (6.25) is referred to as the
quasi-harmonic equation. For three-dimensional heat flow and using the conservation principle together
To solve the differential equation ( 6.24 ), boundary conditions are required. T hese with Gauss' divergence theorem for three dimensions ( cf. (5.32)) the reader may easily
boundary conditions are typically of the form verify that the balance equation (6.22) is again obtained. It then follows directly that
qn = qTn = h on !t;. (6.29)
Strong form of three-dimensional heat flow
T =g on 2g ( 6.30)
div(D VT) + Q= 0 in region V (6.35)
where h and g are known quantities. ;t;. is that part of the boundary fi' on which
the flux qn is known, whereas 2g is that part of the boundary fi' on which the on surface sh (6.36)
temperature Tis known. The sum of the boundaries ;t;. and 2g constitutes the entire
on surface S g ( 6.37)
boundary !t' ( cf. Figure 6.5 ).
84 Introduction to the finite element method Strong and weak forms- 2-D an d 3-D heat flow 88
where S h and S g are those parts of the boundary S on which the• flux q" and 6.3 Weak form of heat flow in two and three dimensions
temperature T are known, respectively. The sum of the boundanes Sh and S9
constitutes the entire boundary S. As an example of the differential equation ( 6.35) We could derive the weak forms directly from the differential equations (6.32) and
assume that D is given by ( 6.11 ). We then obtain the quasi-harmonic equation in three (6.35). However, it is more instructive first to derive the weak form of the balance
dimensions: equation given by (6.21) and (6.22) for two- and three-dimensional problems,
respectively.

ax + !__(
axa ( kx;;c aT) ay kyy aT) az zz aT)+
az Q = 0 ( 6.38) Considering two-dimensional problems, we therefore multiply (6.21) by an
ay + !_(k arbitrary function v - the weight function - and integrate over the region considered,
i.e.
•
For isotropic materials we conclude that

-a ( k -aT) +-
a ( aT) a ( aT)
L v div(qt) dA- L vQt dA = 0 (6.42)

ax ax ay kay- + -az kaz- +Q=O

( 6.39)
where v = v(x, y). Integrating by parts the first term using the Green-Gauss theorem
of ( 5.26 ), we obtain ·
Furthermore, if the thermal conductivity k is constant, (6.39) reduces to

( 6.40)
L v div(qt) dA = f!l' vtqTn d..<t'- L(Vv)Tqt dA (6.43)

Expression (6.42) can then be rewritten in the form

i.e. the Poisson equation in three dimensions. If the heat input Q = 0, ( 6.40) reduces to
L (Vv)Tqt dA = f!l' vqTnt d..<t'- t vQt dA (6.44)

( 6.41) This is the weak form of the balance principle. We may use the boundary condition
(6.33) to obtain
i.e. the Laplace equation in three dimensions.
With these observations we may summarize the equations for three-dimensional
heat flow as shown in Figure 6.6.
fA
(Vv)Tqt dA = f .!ti,
vht d..<t' + f~
vqnt d..<t'- f A
vQt dA (6.45)

where h is a known quantity along ..<e,, whereas the flux qn is unknown along the
boundary~· Expression (6.45) is the weak form of the balance equation (6.21).
Insertion of the constitutive relation (6.4) in (6.45) provides the following weak
form:

Weak form of two-dimensional heat flow

oun ary

+ -f f f
constitutive conditions
relation:
(Fourier)
q = -DVT
qn = h on sh
T = g on S fA
(Vv)Tt D VT dA

T= g
=
.!ti,

on~
vht d..<t' -
, ffg
vqnt d..<t' +
A
vQt dA (6.46)

(6.47)

temperature
T
where v is any function. This result may also be derived directly from the strong form
( 6.32)-( 6.34 ), but the advantage of first establishing the weak form of the balance
Figure 6.6 Illustration of fundamental equations of three-dimensional heat flow principle (cf. (6.45)) and then introducing the constitutive relation (6.4) is that the
86 Introduction to the finite element method
Strong and weak forms- 2-D and 3-D heat flow 87
assumptions behind the different steps become more apparent. Whereas the balance
equation (6.21) or (6.45) is a fundamental physical principle, the constitutive relation
6.4 Other analogous physical problems
( 6.4) is an assumption. In principle, constitutive relations other than Fourier's law
may be envisioned, but the balance principle always holds. We note that if the Just as for one-dimensional heat flow, it follows directly that the formulations for
thickness tis constant, it may be cancelled in (6.46). two- and three-dimensional heat flow apply to the djffusion of ions and to electric
The reader may verify that the weak formulation of the balance principle (6.22) curre~ts. T~is is a~hieve? by generalizing Fick's law in one dimension, ( 4.11 }, and
in three dimensions is Ohms law m one dtmenswn, ( 4.12), to two and three dimensions, as in the generalized
form of Fourier's Jaw (6.4).
• We ~ave already ~entioned that the Poisson equation in two dimensions, (6.27),
( (Vv)TqdV= ( vhdS+f vqndS- ( vQdV (6.48) also apphes to the torsiOn of non-circular elastic shafts and this will be demonstrated
Jv Js. s. Jv in Chapter 14.
Another important example, which also leads to. a formulation similar to that
where the Green-Gauss theorem of (5.33) has been used. In (6.48), the arbitrary of heat flow, is the flow of fluid through porous media (Bear, 1979). To demonstrate
weight function v depends on all the coordinates, i.e. v = v(x, y, z). Introducing the this, consider water flowing in the xy-plane of Figure 6.7, which shows a section
constitutive relation ( 6.4) in ( 6.48) we obtain the following: of the underground.
A tube is drilled into the ground until point A is reached. At point A the water
pressure ~s PA and the. water will therefore rise inside the tube to a level PA f (pg)
above pomt A. Here p 1s the mass density of the water and g is the acceleration due
Weak form of three-dimensional heat flow to gravit~. We ~hoose. an arbitrary horizontal datum level from which the height z is
defined, 1.e. pomt A 1s located zA above the datum level. The same procedure is
( (VvlDVTdV= - f vhdS- ( vqndS+ ( vQdV (6.49) adopted for a tube drilled down until point B. An instrument used to measure the
Jv s. Js. Jv pressures PA and p8 as described above is called a piezometer.
T = g on S 9 (6.50) At an arbitrary point the so-called piezometric head ¢ is defined as

c/J=z+!!_ ( 6.51)
pg

where v is any function. Similar to the one-dimensional case, the boundary condition It appears that¢= ¢(x, y) has the dimension of length and for the case considered
prescribing the derivative of the unknown (i.e. the flux (6.33) and (6.36)) is called a
natural or Neumann boundary condition since it emerges directly from the weak
formulation. The boundary conditions prescribing the unknown itself (i.e. ( 6.34) and
(6.37)) is called an essential or Dirichlet boundary condition. We also note that the
convection boundary condition given by ( 6.31) is occasionally termed a Newton
boundary condition. Boundary conditions like (6.33) and (6.34) which contain both
natural and essential conditions are also called mixed boundary conditions.
PA y
The weak forms above have been derived from the strong forms. As in the

L,
pg
derivation in Chapter 4, it is rather easy to prove that the weak forms also imply A
the strong forms, i.e. the weak and strong forms are identical. The interested reader
may consult Hughes ( 1987). However, from the discussion in Chapter 4 we recall the 4>A
advantages of the weak form, namely discontinuities are allowed for and the order
of differentiation of the unknown function T has decreased. This decrease has been
ZA ~8 8
obtained at the expense of differentiation of the arbitrary weight function v. We also z
observe that this implies that the order of differentiation for v and T is the same.
This indicates a certain symmetry, which is of importance when the FE method is _____ j __ Datum l eve l
formulated.
Figure 6.7 Flow of water underground; piezometric head cp
88 Introduction to the finite element method Strong and weak forms - 2-D and 3-D heat flow 89
Table 6.1 Examples of the heat equation dimensional flow is
Differential equation Physical problem Quantities Constitutive law
q = -k d¢ (6.52)
div(DVT) +Q= 0 Heat flow T = temperature Fourier dx
D = constitutive matrix for q = -DVT
for thermal q = heat flux vector where q is the flow rate, i.e. the volume of water passing through a unit area per unit
conductivities time. It appears that the parameter k, the coefficient of permeability, has the dimension
Q = heat supply [m/ s]. Moreover, Darcy's law (6.52) is quite analogous with Fourier's law (4.3) for
div(DVc) + Q = 0 Diffusion c = ion concentration Fick heat flow, Fick's law (4.11) for ion diffusion and Ohm's law (4.12) for electric currents.
D = constitutive matrix for q = - DVc Just like these constitutive relations (see ( 6.4)) Darcy's law (6.52) can be generalized to
diffusion coefficients q = ion flux vector

div(DVt/>) + Q = 0 Fluid flow

Q = ion supply
4> = piezometric head Darcy
I q= -DV¢ (6.53)
through porous D = constitutive matrix for q = -DVt/>
q = volume flux vector
If D, for instance, is given by (6.11), then as in (6.25) the two-dimensional problem
media permeability
coefficients of Figure 6.7 leads to the following differential equation:
Q = fluid supply
div(DVV) + Q= 0 Electric currents V =voltage Ohm
!_(tk
ax XX
a¢)+
ax !_(tk a¢)+ tQ = o
ay ay
yy
(6.54)
in extended D = constitutive matrix for q = -DVV
bodies electric conductivities q = electric charge flux where Q now denotes the water volume supply per unit volume of the body and per
Q = electric charge supply vector
unit time and t is the thickness of the section underground considered.
div(V¢) + Q = 0 Irrotational, ¢ = velocity potential v = - V¢ Finally, we mention that the so-called irrotational flow of an incompressible and
frictionless and Q = fluid supply v = velocity (not a frictionless fluid is also expressed by the equations given above (see Hughes and
incompressible constitutive law)
Brighton, 1967). In this case the velocity vector v is given as v = - V¢ where ¢ is
fluid flow
the so-called velocity potential. It follows that ¢ satisfies the Poisson equation given
div( ~ V¢) + 211 = 0 Saint-Venant
torsion of elastic
¢ = t/>(x, y) = Prandtl's
stress function
Hooke
U xz= Gyxz
by (6.27) and (6.40) for two- and three-dimensional flow, respectively.
We have seen that a variety of important physical phenomena leads to the same
bars G = shear modulus
II = rate of twist
u,. = Gy,.
Uxz = otjl joy; U 1 z = -otjl j ox
differential equation and boundary conditions as the heat flow problem, and some
examples are collected in Table 6.1. It is with this perspective in mind that we shall
div(Vw) +f = 0 Small deflections w = w(x, y) = deflection
evaluate the significance of the FE formulation for the heat flow problem and which
s of membrane p = lateral pressure will be treated in later chapters.
S = constant tension force
per unit length

we have ¢A> ¢ 8 . Occasionally¢ is called the hydraulic potential. The term p f(pg)
is called the pressure head.
Let us imagine that a tube just containing water is connected between points A
and B. It is obvious that water will then flow from point A towards point B even
though the pressure p8 is greater than the pressure PA· Therefore, it is not the
difference in pressure p that creates the water flow. What really forces the water to
flow from point A towards point B is the fact that ¢A> ¢ 8 . This difference in
piezometric head ¢ is the driving force for the water flow even when the imaginary
water-filled tube between A and B is removed.
From these arguments we are led to Darcy's law from 1856, which for one-
 Approximating functions - scalar problems 91
7 and simpler parts and then, based on these results, it is established for the entire
regwn.
Choice of approximating The first route (a) does not present an attractive general approach since the
region differs from problem to problem. Therefore, this approach is geared towards
functions for the FE the solution of specific problems having in principle the same geometry. What we
are looking for is a method applicable to general geometries and this objective is
method - scalar problems achieved by the second alternative, route (b), described above.
If we split the body, i.e. the region of interest, into smaller parts, i.e. finite elements,
for which we can establish an approximation for the unknown temperature T, then
we are able to analyze bodies with arbitrary geometries. As already touched upon
in Chapter 1, it is a characteristic feature of the FE method that the approximation
of the unknown function is carried out ' elementwise'. Actually, the FE method owes
its name to this characteristic feature.
We have emphasized that the FE method is based on the weak formulation of the In this chapter we shall treat the approximation of a scalar function like the
problem. As a prototype of such a weak formulation we may consider (6.46) for temperature over one-, two- and three-dimensional elements and we shall show how
two-dimensional heat flow: the approximation of the scalar function over the entire region of the body of interest

IA
(Vv)TtD VT dA = -I .!l1,
vht d2- I
.$fg
vqnt d2 +I
A
vQt dA (7.1)
can be established based on this elementwise approximation.
In order to achieve such an approximation we have to make a choice for the
type of approximation. For instance, should trigonometric or exponential functions
In this equation, terms involving the unknown temperature T and the arbitrary be used in the approximation procedure? It turns out that the use of polynomials is
weight function v appear. Therefore, in order to achieve an FE formulation from especially advantageous and convenient for establishing an elementwise approximation
(7.1 ), we need to make certain choices for the approximation of the temperature T of the unknown scalar, and in order to understand this important point, it is
as well as for the weight function v. appropriate to evaluate the criteria for convergence.
We can therefore summarize the basic steps in the FE formulation as follows:

Basic steps in the FE formulation 7.1 General requirements

1. Establish the strong formulation of the problem.
2. Obtain the weak form of the problem. 7.1.1 Convergence criteria
3. Choose approximations for the unknown function - in the present
case the unknown temperature T. The FE method provides an approximate solution to the problem at hand and it is
4. Choose the weight function v. obvious that the more elements we use (i.e. the more computer power we utilize) the
more accurate the approximate solution. In the limit, when the elements are infinitely
small, we require that our approximate solution is infinitely close to the exact
solution. This is the convergence requirement.
Considerable emphasis has already been given to steps 1 and 2. In the present chapter
Let us consider the inhomogeneous temperature field over the two-dimensional
we shall treat the manner in which approximations are made for the unknown
body shown in Figure 7.1. The boundary along AC is insulated and the temperature
function T, i.e. step 3. In the following chapter we shall finally make an evaluation
is prescribed along the boundary ABDC. In Figure 7.1(b), a very coarse element
of a suitable choice of the weight function v, i.e. step 4. After these discussions we
mesh consisting of only two triangular elements is used, whereas many triangular
will eventually be in a position to present the FE method.
elements are used in the fine element mesh shown in Figure 7.1 (c).
In order to establish an approximation for the unknown scalar- the temperature
Within each element the temperature varies drastically in the coarse mesh and
T - we may in principle adopt two basically different routes: (a) the unknown
less dramatically in the fine mesh. In the limit, for infinitely small elements, we then
function is approximated directly over the entire region of interest; (b) the region is
have within each element either a constant temperature gradient or a constant
divided into smaller parts and the approximation is first carried out over these smaller
temperature. If the approximation of the temperature over each element is to

90
92 Approximating functions- scalar p r oblems 93
Introduction to the finite element method

al T= 100 °C B

Insulated
Figure 7.2 Two neighbouring elements

T= oc a ion~ the dotted line. We are then led to the so-called compatibility or conforming
50 reqwrement:

bl cl
Compatibility or conforming requirement
• The approximation of the temperature over element boundaries must
be continuous.

The convergence requirement, stating that for infinitely small elements the
Figure 7. 1 (a) Two-dimensional heat problem; (b ) coarse element mesh; (c) fine element mesh approximation approaches the exact temperature field, is clearly fulfilled if both the
completeness and compatibility requirements are fulfilled, i.e. we obtain the important
statement
approach the exact temperature for decreasing element size, it follows that the
approximation must fulfil the so-called completeness requirements : convergence criterion =completeness +compatibility requirements

Here only qualitative arguments have been presented for the fulfilment of
Completeness conv~rgence, but a more stringent evaluation is presented by Strang and Fix {1973 ),
and m Chapter 9 we shall rederive the compatibility requirement using a different
• The approximation must be able to represent an arbitrary constant
approach. However, it turns out that it may even be allowable to relax the
temperature gradient.
compatibility (i.e. the conforming) requirement and still be able to fulfil the
• The approximation must be able to represent an arbitrary constant
convergence criterion. Indeed, even if a discontinuity exists across an element
temperature.
boundary, this discontinuity may approach zero for infinitely small elements. Finite
elements fulfilling the compatibility (i.e. the conforming) requirement are called
conforming elements, otherwise they are termed non-conforming elements. Most finite
The completeness requirements therefore imply that the approximation of the
elements are conforming, but even non-conforming elements are used, especially in
temperature at least must include terms corresponding to an arbitrary linear
plate analysis as discussed in Chapter 18.
polynomial. That is, in three dimensions we must have the following approximation:
As indicated above, the fulfilment of both completeness and compatibility is
sufficient for convergence. However, whereas the requirement of compatibility may
(7.2)
be relaxed, the fulfilment of completeness is clearly a necessary condition for
convergence that must be satisfied.
Consider now two neighbouring elements as shown in Figure 7.2, where
triangular elements are used for convenience. The temperature variation along the As the completeness requirement must be met, one element is always able to
reproduce exactly an arbitrary linear temperature variation. Clearly, if an arbitrary
common element boundary AB is considered. In each element an approximation of
assemblage of conforming elements is considered, these elements are also able to
the temperature is made and if the temperature variation in element 1 along the
dashed line is evaluated, it may differ from the temperature variation in element 2 reproduce exactly an arbitrary linear temperature variation over the assemblage. This
94 Introduction to the finite element method Approximating functions- scalar problems 95
observation leads to the so-called patch test for non-conforming elements. To check written as
that non-conforming elements fulfil the convergence requirement they must pass this
test. The essence of the patch test is to check that an arbitrary assemblage (i.e. patch) T(x, y) = T0 + (oT) (x- x0 ) + (oT) (y- y0 )
of elements is able to reproduce exactly an arbitrary linear temperature variation. ox o oy o
Apart from some plate elements discussed in Chapter 18, in this introductory text
+ l(~::)o (x- Xo) + (o~~Y)o (x- x y
2
we shall only consider conforming elements and we refer to Hughes ( 1987), Strang 0 )(y- 0) (7.3)
and Fix (1973) and Zienkiewicz and Taylor (1989) for a detailed treatment of the
patch test and the behaviour of non-conforming elements.
Let us return to the completeness requirements which must be fulfilled in order
to achieve convergence. Referring to (7.2) we note that an approximation which at
+ l (~::) 0
(y- y 0 )
2
+ O(r 3 )

least involves an arbitrary linear polynomial is necessary in order to fulfil the where the term O(r 3 ), which is read as 'of order 3', is defined in general as
completeness criterion. With this result at hand it will come as no surprise to see O(r") = Cr" for r ~ 0 (7.4)
later that the elementwise approximations always consist of polynomials possibly
containing even higher-order terms than linear ones. In this expression, C is a constant and r is the distance between the current point
(x, y) and the fixed point (x0 , y 0 ) (cf. Figure 7.3). Moreover, in (7.3) T0 is the exact
temperature at point (x 0 , y 0 ) and, likewise, (oTf ox)0 , (oTf oy)0 , (o 2 Tfox 2 ) 0 ,
(o 2 T/ox oy)0 and (o 2 T /oy 2) 0 are the exact derivatives evaluated at the point (x 0 , y 0 ).
It appears that (7.3) may be written as
7.1.2 Rate of convergence- Pascal's triangle
(7.5)
Having established that the approximation of the unknown function (temperature) where a 1 ... a6 are constants.
is obtained using a polynomial, which at least includes all linear terms, we need In a region which contains the point (x0 , y 0 ) we now assume that the
information on which higher-order terms it may be advantageous to include. For approximation to the true temperature field is taken as
this purpose we shall discuss the concept of rate of convergence, which provides
information on how fast the approximate solution converges towards the exact T"PP(x, y) = a 1 + a 2 x + a 3 y + a 4 x 2 + a 5 xy + a 6 y 2 (7.6)
solution when the element size is decreased. where the notation T"PP has been used in order to emphasize that (7.6) represents
Consider the two-dimensional body shown in Figure 7.3. Assume that the exact the approximate temperature field. Subtracting (7.6) from (7.5) yields
temperature distribution T is known and let us make a Taylor expansion of this
exact solution about the point (x 0 , y 0 ). It follows that the true solution T can be T(x, y)- T"PP(x, y) = Cr3 (7.7)
and it appears that T - PPP is the error in the approximation process. Let us now
consider a finite element (i.e. a region) which contains the point (x0 , y0 ) and let h
denote the maximum distance from the point (x0 , y 0 ) to the element boundaries.
y Within this finite element the maximum error is written as IT- T"PPImax and from
(7.7) we find that
IT- T"PP imax = Ch 3
We may consider the distance h as representative of the element size and to emphasize
this issue, the result above is written as
(7.8)
where T~PP refers to the approximation for an element of size h.
Let us assume now that instead of an element of size h we use another,
L------------------------------x geometrically similar element of size rxh. In analogy to (7.8) it follows that
Figure 7.3 Taylor expansion of true solution about point (x0 , y 0 ) IT- T~fPimax = Crx 3 h 3 (7.9)
96 Introduction to the hnite element method Approximating functions - scalar problems 97
Combining (7.8) and (7.9) results in more accurate than (7.12), owing to the term a 5xy, and this is reflected by the fact
IT - r:~p Imax 3
= o: 1T - ThPP Imax (7.10)
that, in general, C 2 ~ C 1 (cf. (7.13) and (7.17)). However, the important point is that
the rate of convergence is unchanged. Terms in the approximation which do not
If, for example, a is chosen as t, i.e. the new element size is half its original size, we improve the rate of convergence are called parasitic terms.
obtain Obviously, unless other specific properties of the approximating function are
required, one would try to avoid parasitic terms and instead use polynomials whlch
(7.11)
contain terms of lower order before hlgher-order terms are invoked. A polynomial
Thls implies that with an element size of h/ 2, the maximum error in the whlch contains all terms of a specific order as well as all lower-order terms is called
approximation is one-eighth of the maximum error using an element size of h. The a complete polynomial (not to be confused with the completeness requirements,
term o:3 present in (7.10) gives the rate of convergence, whlch in the present case is discussed previously). In one dimension, the approximations a 1 + a 2 x and
of order 3. It follows that the rate of convergence provides information on how fast a 1 + a2 x +a 3 x 2 are complete, whereas the approximation a 1 + a2 x + a 3 x 3 is
the approximate solution converges towards the true solution when the element size incomplete because it lacks the quadratic term.
is decreased. Observe that as the exact solution T is unknown in practice, it is not In order to keep track of the terms in a polynomial used for two-dimensional
possible to calculate the value of the error IT - PPP imax· approximation it is convenient to make use of Pascal's triangle, whlch lists the terms
In (7.6) we considered all terms up to and including the quadratic terms. Suppose in a systematic manner:
now that only the linear terms are considered in the approximation, i.e.
yapp(X, y) = al + a2X + a3y (7.12) Pascal's triangle
As, for instance, the term (o 2T/ ox2)0(x- x 0)2 1s of order O(r2 ) we obtain by l
subtracting (7.12) from (7.3) that
X y
(7.13) x2 xy y2
where C 1 is a constant. As in the derivation of (7.10), we obtain from (7.13) that x3 x2y xy2 YJ
x4 x3y x2y2 xyJ y4
(7.14)
i.e. the rate of convergence is of order 2 for the approximation given by (7.12). If the
element size is halved, i.e. a= t, (7.14) yields
IT - T h/appl - ~ ~T - T"PPI Any polynomial which contains all the terms above a certain horizontal line in
2 max - 4 h max (7.15)
Pascal's triangle is a complete polynomial. For one-dimensional problems Pascal's
A comparison with (7.11) illustrates that the more higher-order terms we include in
triangle degenerates to 1, x, x 2 , x 3, ... and even for three-dimensional problems it is
the approximation, the faster is the rate of convergence.
rather straightforward to generalize the results above.
However, this conclusion should be used with caution as the following example
With these results in mind, we can summarize our discussion for the elementwise
will show. Consider the approximation given by
approximation procedure:
(7.16)
Even though we include one quadratic term, the true solution (7.3) contains other
quadratic terms which are of the order O(r 2 ). That is, subtracting (7.16) from (7.3)
Elementwise approximation
yields • A polynomial is used.
• Thls polynomial must include at least an arbitrary linear polynomial -
(7.17) then the completeness requirements are fulfilled.
where C 2 is a constant. A comparison of (7.13) and (7.17) shows that the two • Choose the polynomial so that continuity across element boundaries is
expressions are similar in structure even though C 1 "# C 2 . Thls means that achleved - then ttie compatibility or conforming requirement is fulfilled.
approximation (7.16) also leads to (7.14) just like approximation (7.12), i.e. we have • If possible, avoid parasitic terms.
not improved the rate of convergence. The approximation given by (7.16) is in general
98 Introduction to the finite element method Approximating functions - scalar problems 99
Having established these general guidelines, we shall now discuss the elementwise temperatures and 7; ~ at the nodal points i and j, respectively. From (7.18) and
approximation of a scalar function like the temperature for one-, two- and Figure 7.4 we obtain
three-dimensional problems, and we will show how the approximation of the scalar
function over the entire region of the body of interest can be established based on
these elementwise approximations. For convenience the scalar function is denoted
by T, as it may represent the temperature of a body.
7; = + a2xi
et1

~ = a 1 + a 2 xi or [~] = [~ ::J[:J
which can be written as
a• = c~ (7.21)

7.2 One-dimensional elements where

7 .2.1 Simplest possible one-dimensional element a• [7;]; C [11 xi]

=
~xi
= (7.22)

Referring to (7.2), it appears that the simplest possible one-dimensional approxima- It appears that the column matrix a• contains the temperature at the nodal points
tion, which fulfils the completeness requirements, is given by of the element. From (7.21) it follows that
(7.18) (7.23)
where a 1 and a 2 are constant parameters for each element. In a moment we will and insertion into (7.19) yields
express these parameters in terms of the temperature at the nodal points of the
element. As we have two parameters, the element contains two nodal points.
T = NC- 1a• (7.24)
Moreover, it is advantageous to locate the nodal points at the ends of the element, This expression can be written as
i.e. we obtain the element with the length L as shown in Figure 7.4.
Let us rewrite (7.18) in the following matrix form T= N•a• (7.25)
T=N~ (7.19)
where
where
N• = [Nf Nj] = NC- 1 (7.26)

N = [1 x]; ~ = [::] (7.20) From (2.46) it is easy to determine the inverse matrix c- 1 and we obtain
It turns out to be advantageous to express the parameters a 1 and a 2 in terms of the c-1=2.[
L
xj
-1
-X·]
1
1
whereL =x . -x.
J I
(7.27)

Using this expression in (7.26) it follows that

T
I N· = [Nf Nj] (7.28)

where

1
N~ = - - (x- X·)
' L J
(7.29)
L_--~========3------- x
I· ·I
Figure 7.4 Simplest possible one-dimensional element The functions Ni and Nj are called element shape functions or element interpolation
100 Introduction to the hnite element method Approximating functions- scalar problems 101

T T

T;

N~lxl
>-'·x /< N~ lxl
"".
'"-.;"' I
N~T.
I

I
/
/

L_--~======~~----- x
/ ·-.......
.
..........
J

L_-c====~>-- x
"'"- L_--e====:::J>-- X

Figure 7.5 Variation of element shape functions with position x

T
functions and the matrix N• is termed the element shape function matrix. It will turn
out later that different methods exist to establish the element shape functions and T·I ~T =NfT; .N~ T.
the specific method we have adopted here is termed the C-matrix method (cf. (7.21)).
~ II
T·J
It is of interest to evaluate how the element shape functions Nf(x) and Nj(x)
vary with position x, and Figure 7.5 shows this dependence. From (7.29) it is obvious
that Ni(x) and Nj(x) vary linearly with position x. Moreover, from (7.29) and Figure
7.5 it follows that the element shape functions possess the following important
properties:
Figure 7.6 Approximation of temperature in linear element
Ni(xi) = 1; N'[(xi) = 0
(7.30)
Nj(xi) = 0; Nj(xi) = 1
derived directly from (7.31 ). As an example, for x =xi we have T = 7;, i.e.
These properties will also turn out to be valid for the more complex elements to be
[N't(xd- 1] 7; + Nj(xJ1} = 0
discussed later.
We note that (7.25) can be written as As this expression holds for arbitrary 7;- and 1)-values, we conclude that Nf(xi) = 1
and Nj(xi) = 0, in accordance with (7.30). Likewise, if x =xi is used in (7.31) we
T(x ) = N'[(x )1; + Nj(x)1j (7.31) conclude that Ni(xi ) = 0 and Nj(x1) = 1, also in accordance with (7.30).
Referring to the weak formulation ( 4.36) of the one-dimensional heat flow
With expressions (7.25) and (7.31) we have obtained a formulation which separates problem, it is important to determine the gradient of the temperature T. From (7.25)
the influence of geometry via the element shape functions from the influence of physics or (7.31) it follows that
via the temperatures ·at the nodal points. This appealing separation is not present in
dT dN• dNi dN•
(7.18), where the parameters a 1 and cx2 depend on geometry as well as on nodal point - = - a·= - 7 ; + -1 1j (7.32)
temperatures. It follows that once the geometry of the element is known, the element dx dx dx dx
shape functions can be established directly. This important feature is common for Let us define the matrix s• by
all types of finite elements and it greatly facilitates implementation on a computer.
Expression (7.31) shows clearly that the approximate temperature is expressed
as a suitable interpolation between the temperatures at the nodal points. This
~ (7.33)
interpolation is given by the element shape functions, which in the present case are
linear functions. Moreover, the approximate temperature T depends linearly on the
~
Then (7.32) can be written as
temperatures 7; and 1j at the nodal points, i.e. the approximation of the temperature
can be illustrated as shown in Figure 7.6.
When the fundamental properties (7.30) for the element shape functions were ~ (7.34)
established, use was made of the expressions given by (7.29). However, (7.30) can be
~
100 Introduction to the finite element method
Approximating functions- scalar problems 101
N~ N!
I ' J T T

T·I
"
'"~I
N~T.
I
L---~========~------x
"'"·
L....----{:======3---- X

Figure 7.5 Variation of element shape functions with position x

T
functions and the matrix N• is termed the element shape function matrix. It will turn
out later that different methods exist to establish the element shape functions and T·I
the specific method we have adopted here is termed the C-matrix method (cf. (7.21 )).
It is of interest to evaluate how the element shape functions N~(x) and Nj(x) T·J
vary with position x, and Figure 7.5 shows this dependence. From (7.29) it is obvious
that Ni(x) and Nj(x) vary linearly with position x . Moreover, from (7.29) and Figure
7.5 it follows that the element shape functions possess the following important L----£====:::9---- X

properties:
Figure 7.6 Approximation of temperature in linear element
Nf(x1)= 1; Nf(xj)=O
(7.30)
Nj(x 1) = 0; Nj(x1) =1
derived directly from (7.31). As an example, for x = x 1 we have T = T;, i.e.
These properties will also turn out to be valid for the more complex elements to be
discussed later.
[Nf(xl) - 1]T; + Nj(x 1)1j = 0
We note that (7.25) can be written as As this expression holds for arbitrary T;- and T1 values, we conclude that Nr{x1) = 1
and Nj(xJ = 0, in accordance with (7.30). Likewise, if x = x 1 is used in (7.31 ) we
T(x) = N~ (x)T; + Nj(x)1j (7.31) conclude that Ni(x) = 0 and Nj(x1) = 1, also in accordance with (7.30).
Referring to the weak formulation ( 4.36) of the one-dimensional heat flow
With expressions (7.25) and (7.31) we have obtained a formulation which separates problem, it is important to determine the gradient of the temperature T. From (7.25)
the influence of geometry via the element shape functions from the influence of physics or (7.31) it follows that
via the temperatures at the nodal points. This appealing separation is not present in
dT dN• dN~ dN•
(7.18), where the parameters a 1 and a2 depend on geometry as well as on nodal point - =-a• =--~ T; + _ J T. (7.32)
temperatures. It follows that once the geometry of the element is known, the element dx dx dx dx J

shape functions can be established directly. This important feature is common for Let us define the matrix s• by
all types of finite elements and it greatly facilitates implementation on a computer.
Expression (7.31) shows clearly that the approximate temperature is expressed
as a suitable interpolation between the temperatures at the nodal points. This ~ (7.33)
interpolation is given by the element shape functions, which in the present case are
linear functions. Moreover, the approximate temperature T depends linearly on the
~
Then (7.32) can be written as
temperatures T; and 1j at the nodal points, i.e. the approximation of the temperature

~
can be illustrated as shown in Figure 7.6.
When the fundamental properties (7.30) for the element shape functions were
~
(7.34)
established, use was made of the expressions given by (7.29). However, (7.30) can be
102 Introduction to the finite element method Approximating functions - scalar problems 103

Therefore the matrix B• relates the temperature gradient to the temperatures at the the element. We note that, due to (7.26) and (7.20), B• may be written as
nodal points of the element. From (7.33) and (7.29) we obtain
B• = dN• = dN c - 1 = [O 1]c- l (7.36)
dx dx
B• = [ d~i J ±±J
dd:j = [ _ (7.35) where c-l is given by (7.27).
We have already mentioned that the approximation chosen fulfils the complete-
ness requirements. Let us now check that the element formulation also fulfils the
As B• here is a constant matrix, the gradient dTjdx is constant within the element
compatibility or conforming requirement stating that the temperature approximation
in accordance with the fact that the temperature is assumed to vary linearly within
must vary continuously across common element boundaries. For this purpose,
consider the body consisting of three elements shown in Figure 7.7, where, for
instance, NJ is the element shape function of nodal point j for element 1, wh~reas
T NJ is the element shape function of nodal point j for element 2. For each element,
T
we have the approximation shown in Figure 7.6 and as we interpolate between the
J
in element 1 nodal points, and as, for instance, nodal point j is common for two elements, we
T·I clearly have a continuous variation of the temperature over common element
boundaries as shown in Figure 7.7. This means that the element is conforming.
However, we note that the temperature gradient, in general, varies discontinuously
over common element boundaries.
Figure 7.7 shows how the approximate temperature variation over the entire
T
body is established from the approximation over each element. We shall now
in element 2 formulate the approximate temperature variation over the entire body in a slightly

~ different manner. Previously, we determined the element shape functions (cf. (7.29)
and Figure 7.7) from which it appears that one element shape function is related to
+ N2j Tj ..--/ ' v'
/---.:... N T
/
2
k k
nodal points i and m, whereas two element shape functions are related both to nodal
/ "'-.. point j and nodal point k. Let us now define the global shape functions in such a
L_~c====L======~====~---x manner that one global shape function is related to each nodal point. The following
CD ® k ® m 'definitions are adopted:
T 1
N- = {N; for x in element 1
·' 0 otherwise

+
r
for x in element 1
Ni = NJ for x in element 2
0 otherwise
(7.37)

r
for x in element 2
T
Nk = :~ for x in element 3
over entire body otherwise

Nm = {N~
0
for x in element 3
otherwise
These global shape functions as well as definitions (7.37) are illustrated in Figure 7.8.
In this figure, it appears that the global shape function for a: specific nodal point
Figure 7.7 Approximation of the temperature over three elements only differs from zero in those elements which contain this nodal point. This feature
104 Introduction to the finite element method Approximating functions- scalar problems 105
ELEMENT SHAPE FUNCTIONS GLOBAL SHAPE FUNCTI ONS T

'L'~·: § § 0 X
N.I

X
+ T;L, : '<:"NiTi
§
§ 0 ' X
CD @ k ® m CD @ k m
T
Ni
T·J
N!-.... /......._, .....---N~ /',
J
/
y .........
-......... J
+ /
'
1- N · T· ~'
X X I J J "-
CD ® k ® m X
m

./ ...-...._, _.......N~
=> Nk

Tk
T

"- X X + / """ '

//..._Nkrk....-',
CD
X
Nm
k m

r.L
N3 T
m........_,.. /
/

CD j @ k Q) m
X
(j) @ k Q) m
X
+ NmTm-.....-
.:./
Figure 7.8 Definition of global shape functions from element shape functions § § 9 X
k m
will turn out to be of fundamental importance in the later FE formulation and it is T
characteristic of all types of finite elements.
With these definitions of the global shape functions, it is obvious that the
approximation of the temperature over the entire body can be written as
(7.38)
as illustrated in Figure 7.9. The reader is encouraged to compare Figures 7.7 and 7.9 k m
and check that (7.38) holds. Figure 7.9 Approximation of the temperature over entire body using global shape functions
The temperatures at all nodal points of the body are collected into the column
matrix a given by Then (7.38) can be written as

·-[~]
T=Na (7.41)
(7.39)
It follows that the gradient within the body is given by

Moreover, the global shape function matrix N is defined by ~ (7.42)

N = [N; Ni Nt NmJ (7.40) ~
 Approximati ng func tions - scalar problems 107
106 Introduction to the finite element method

where B is defined as Let us rewrite (7.45) according to

T= Ncx (7.46)
(7.43) where

i.e. (7.47)

B= [dN; dNJ dNk dNm] (7.44)

dx dx dx dx
In order to express the parameters a 1 , a 2 and a 3 in terms of the temperatures at the
As these results are generalized directly to bodies comprising more elements, we have
nodal points, we express that (7.45) for x = x 1 provides T = T;, etc. This yields
obtained a complete description of the approximate temperature over an arbitrary
one-dimensional body. Note that the elements may be of different lengths. 7'; = oc 1 + a2X1 + a3 Xf
~= a1 + a 2 x1 + a 3 xJ
7.2.2 Quadratic one-dimensional element 7',. = a 1 + a2 xk + a3x;
which can be written as
Instead of the simple linear one-dimensional element presented above we may easily
a• = Ccx (7.48 )
construct higher-order elements where the approximation of the temperature contains
terms of higher order than the linear ones. As an example consider the quadratic
where
approximation given by
(7.45) X; X~] (7.49)
x1 Xi
Again it is advantageous to express the three parameters a 1, a 2 and a 3 by the
temperature at the nodal points. That is, there is room for three nodal points as xk x;
shown in Figure 7.10. In order to obtain continuity in the temperature across As before, the column matrix a• contains the temperatures at the nodal points of the
neighbouring elements, nodal points are located at each end of the element and the element. From (7.48) it follows that
third nodal point may be located arbitrarily. In practice, however, the third nodal (7.50)
point is located at the middle of the element as shown in Figure 7.10.
and insertion into (7.46) yields
T = NC-'a• (7.51 )
T
This expression can be written as

T = N•a• (7.52)

where
k N• = [ Nf Nj Nk] = NC-
1 (7.53)
L---~==~~==~------
0 x
xi xi xk Let us now determine the inverse matrix c - 1 . First, we determine the determinant
L/2 L/2 of C. From (2.40) and choosing j = 1 we obtain
·I
L
·I det C = 1(x1x;- xkxJ)- l( x 1x;- xkxl) + 1(x;xJ - x1xl>
Figure 7.10 Quadratic one-dimensional element = X j X k( X t - x 1) - X ; Xk ( Xk - X 1) + X 1X 1(x1 - X;)
108 Introduction to the finite element method Approximating functions- ~cc:@r problems 109

Reference to Figure 7.10 gives T T T

det C =-
L L
(x1xk- 2X 1Xk + X1X1) = - [xk(Xj- X;)+ X;(Xj- Xk)]
/
2 2
+ T.
J +
= ~ ( xk ~ - ~) = ~ (xk -
2

=~
3 T-
x1 x;) '
L
X --~~-~-~~==3----x
Using (2.46), after a little algebra we then obtain k i \:_~_:; ·J k

x 1xk - 2x1xk
c-l = :2 [
-(xj + xk) 2(x; + xk) (7.54)
T

1 -2
Inserting (7.47) and (7.54) into (7.53) provides

I Ne = [Ni Nj NU (7.55)
L~==~oc==:9---- x
k
where
Figure 7.12 ApproximatiWJ of temperature in quadratic elep1ent
2
Ni = L 2 (x- x 1)(x- xk)
4 the following fundamental property
Nj = - L 2 (x- x 1)(x- xk) (7.56)

2 N~ = { 1 at nodal point i
(7.57)
Ni = L 2 (x- x 1)(x- x) ' 0 at all other nodal points

As before Nf, Nj, Ni are called element shape functions, whereas N• is termed the in complete analogy to (7.30).
element shape function matrix. Just as for the linear element, the specific manner by Expression (7.52) may be written as
which we have determined the element shape functions is termed the C-matrix method
(cf. (7.48)). It appears that the element shape functions are quadratic functions of x, T = NiT; + Nj ~ + N'k 1k I (7.58)
and Figure 7.11 shows their variation with position x.
From (7.56) and Figure 7.11 it follows that the element shape functions possess which shows that the temperature T depends liqea,rly on the temperatures T;, ~ !!Ud
1k at the nodal points; that is, the approximation of the temperature can be illus~rated
as shown in Figure 7.12. ...
Just as for the linear element, the fundame\'}tal property (7.57) for the element
shape functions may be derived directly from (7,~58). Let us assume that x ,; ~ - -f~r
which T = T;, i.e. ·· · . ·'

[Ni(x;)- l]T; + Nj(x;)~ + Nk(x1)7k = 0

As this expression holds for arbitrary T;-, ~- a~d T,.-values, we conclud~ ~pat
Ni(x;) = 1, Nj(x;) = N'k(x;) = 0 in accordance with (7.57,). The remaining propertjes
k given by (7.57) are obtained in a similar way by setting .x = x1 and x = xk. ·
Figure 7.11 Shape functions for quadratic element In the weak formulation ( 4.36), the temperature gradient appears. In order to
llO Introduction to the finite element method Approximating functions - scalar problems Ill
determine this gradient we use (7.52) and (7.58) to obtain T

-dT = - - a• = dN•
dN•
__• I; + dN~ dN•
dx dx dx
__J 1j + _
dx
_k 1k
dx
As with the linear element, the matrix B• is defined by
(7.59)

N
1
r.
/
Q7- N~Tk 1

'f
-N 1. T1 , '
j \
in el em ent 1

I I I " \
I \

c:]] (7.60)

Tk
L_-~~==~o~::;J::::::Jo[:::::~---- x

T
i'-. . / j k m n
CD ®
i.e. we have
+ Tm
(7.61) Tn

where the matrix B• relates the temperature gradient to the temperatures at the nodal
points of the element. With the element shape functions given by (7.56) it follows that T

over ent i re body

B• __ [dNi dNj -dNk] = -2 [2x - x1. - xk T·I
dx dx dx L2

(7.62) L_-~C::::o~:::]::::::Jo~:::::}---- x
j k m n
It is no surprise that B• and thus the temperature gradient within the element depends
linearly on x since the temperature was assumed to vary quadratically with x ( cf. CD ®
Figure 7. 13 Approximation of temperature over two quadratic elements
(7.45)). We also observe that, due to (7.53) and (7.47), B• may be written in the
following alternative manner:
the global shape functions in such a manner that one global shape function is related
• dN• dR _ 1 to each nodal point. The following definitions are made:
B = - = - C = [0 1 2x] C - 1 (7.63)
dx dx N1 for x in element 1
N; = 0 •
{ otherwise
where the constant matrix c - 1 is given by (7.54).
To illustrate how the temperature variation over the entire body is obtained
from the elementwise approximation in a manner analogous to the approach for the
Nj = 0 ) {N1fo r x in element l
otherwise
linear element, we consider the body shown in Figure 7.13, which is assumed to

r
comprise two elements. for x in element 1
The notation is as before, implying that Nf is the element shape function of
nodal point k for element 1, whereas Nf is the element shape function of nodal point
N,~ :; fo r x in element 2 (7.64)
otherwise
k for element 2. Since we interpolate between the temperatures of the nodal points
and as, for instance, nodal point k is common to two elements, we clearly have a for x in element 2
continuous variation in the temperature over common element boundaries. That is,
Nm = {N'!
0 otherwise
the element is conforming or compatible. However, the temperature gradient varies,
in general, in a discontinuous manner over common element boundaries. for x in element 2
N = {N;
Just as for the linear element and with reference to Figure 7.13, we now define n 0 otherwise
112 Introduction to the finite element method Approximating functions- scalar problems 113
E L EMENT SHAPE FUNCT IONS GLOBAL SHAPE FUNCTIONS element shape function N~(x) according to
N.I

N;(x) = {Ni(x) ifelem~nt ex contains the global nodal point i

(7.66)
0 otherwtse

The reader is in vited to check that (7.64 ) is contained in this definition and that even
(7.37), applicable for the linear element, is contained in the fo rm given by (7.65) and
/ ........ ,_N~ (7.66). In fact, we shall adopt the definitions (7.65) and (7.66) for all types of finite
/ '\ J elements. These expressions imply that a global shape function related to a specific
I \
0 X nodal point differs from zero only in those elements which contain the nodal point
j k m n in question.
Q) Nk It appears that the approximation over the entire body given in Figure 7.13 can
now be written as

(7.67)
k
0

m n
X
~ X

as illustrated in Figure 7.15. The temperatures at all nodal points of the body are
® Nm collected into the column matrix a given by

7;
X 1j
m n m n
a= 11.: (7.68)
® 0
Tm
T,
0 X
j k
Moreover, the global shape fu nction matrix N is defined by
Q) N = [N; Ni Nk Nm Nn] (7.69)
Figure 7. 14 Definition of global shape functions from element shape functions Then (7.67) can be written as

T=Na (7.70)

For the body shown in Figure 7.13, these global shape functions take, as well as The gradient within the body becomes
definitions (7.64 ), the form shown in Figure 7.14.
In order to write expression (7.64) in a more compact fashion, let us denote a ~ (7.71)
~
global nodal point by i. For a given position x, we may identify the associated element
with number ex which contains this position; that is,
where B is defined as
position x =>identification of associated element ex (7.65)
(7.72)
With this observation, the global shape function N 1(x) can be defined from the
 Approximating functions- scalar problems 115

i.e.

(7.73)

It is obvious that this concept generalizes directly to bodies comprising more

elements.

7.2.3 Cubic and quartic one-dimensional elements- Lagrange

in te rpola tion
Previously we established the linear and quadratic elements and we may proceed to
consider the cubic element given by
(7.74)
In order to express the parameters a 1 ... a 4 in terms of the temperatures at the nodal
point, it is possible again to use the C-matrix concept given by (7.21) and (7.48),
which requires that the inverse c- 1 is established. For the linear element, this inverse
is easily obtained, but for the quadratic element the calculations are rather
cumbersome. This obstacle will be much more pronounced when an expression like
(7.74) is considered, but it turns out that a general procedure exists by which the
element shape functions can be written down directly.
To establish this general procedure, we shall first adopt a suitable numbering of
the (local) nodal points belonging to an element. It is emphasized that this local
numbering has no effect on the numbers of the global nodal points. With the nodal
numbers of the linear element given in Figure 7.16(a), (7.29) provides the element
shape functions
1 1
N~ = - -(x- x 2 ); Ni = -(x - x 1 ) (7.75)
L L
Likewise, with the nodal numbers of the quadratic element given in Figure 7.16(b ),

a) b)
x, x2 x, x2 x3
X 0 X

Figure 7.15 Approximation of the temperature over entire body using global shape functions
,. L
2

·I I·
L/2
2

·I·
L/2
3

·I
L
.I
Figure 7.16 (a) Linear element; (b) quadratic element
116 Introduction to the finite element method Approximating functions- scalar problems 117
XI Xz X3 x4 x,
~----~==~o==:Jo[::J~-----x ~----4:==~o==~o==~o[:~-------- x
1 2 3 I. 2 3 I. 5
I· L/3· 1,Lf3,1.u3.1
·I
Figure 7.17 Cubic element Figure 7.19 Quartic element

XI X2

~--~========~---
2
x
--- u, --- uz
Figure 7.20 Linear element applicable for a xially loaded bar

directly by setting

k = 1, 2, . .. , n (7.79)
Figure 7.18 Lagrange's interpolation function
To illustrate the use of Lagrange's interpolation function, we shall re-establish
(7.56) results in the element shape functions N~ and N~ for the linear element shown in Figure 7.16(a).
As the element comprises two nodal points, we haven= 2 and (7.79) then provides
N~ =-..; (x - N~ =-
4
L
x 2)(x- x 3 );
L2
(x - x 1)(x- x 3); N~ = IL k = 1, 2
(7.76)
From (7.78) we get
2
N'3 =2 (x - x 1)(x - x 2)
L
Consider now the cubic element in Figure 7.17. It is required that the element
shape functions allow us to write and as L = x 2 - x 1 , we have rederived (7.75). In a similar way, the reader may verify
that for the quadratic element, i.e. n = 3, use of (7.79) yields the results given by
(7.77) (7.76). Therefore, the element shape functions of the cubic element shown in Figure
i.e. that the temperature is interpolated between its values at the nodal points, but 7.17 are obtained directly from (7.79) by putting n = 4, whereas the element shape
in order to do so the element shape functions must possess the fundamental property functions of the quartic element given by Figure 7.19 emerge for n = 5. This process
given by (7.57). We also know that the element shape functions must be a polynomial may be continued for arbitrary n-values. As the element shape functions for all the
and a polynomial fulfilling (7.57) is given directly by Lagrange's interpolation formula. one-dimensional elements considered may be derived using Lagrange's interpolation
In fact, for n given points this formula provides a polynomial of order n - 1 given by formula, these elements are termed Lagrange elements.
Having discussed one-dimensional elements for heat flow problems in great
n-l (x- x 1)(x - x 2 ) ··· (x - xk _ 1)(x- xk+ 1) ·· · (x - x.) . detail, we finally mention that these elements obviously apply to axially loaded bars
Ik (x)= ,
(xk - x 1)(xk- x 2 ) ··· (xk - xk - d(xk - xk+ l) ··· (xk- x.) as well. In principle, however, one aspect needs a little attention. Whereas the
k = 1, 2, ... , n (7.78) temperature T is a quantity without any direction, the axial displacement u has
direction. When the differential equation (4.18) was derived for an axially loaded
where it should be noted that the term (xk - xk) is not present in the denominator elastic bar, the axial displacement u was considered as positive when it was in the
and, likewise, the term (x - xk) is not present in the numerator. It appears that direction of the x-axis. Taking account of these remarks, the simple linear element
zz-1
(xk) = 1 and z;;- 1(x;) = 0 fork# i, as illustrated in Figure 7.18. Moreover, these applicable for the axial loading of a bar is illustrated in Figure 7.20, where u 1 and
properties are in accordance with the fundamental property for element shape u 2 are the displacements of nodal points 1 and 2, respectively. A similar approach
functions given by (7.57). Therefore, the element shape functions can be constructed holds for higher-order elements.
118 Introduc tion to the finite el em ent method Appr oxima ting functions- scalar problems 119

7.3 Two-dimensional elements T

T·J

I
7 .3.1 Simple triangular element I
I
I
After this detailed discussion of one-dimensional elements, it is timely to consider I
two-dimensional elements. Referring to (7.2), the simplest possible two-dimensional T =a 1 +a 2 x + a 3 y
element which fulfils the completeness requirements is given by I
I
(7.80)
As we have three parameters cx 1, a 2 and cx 3 , which are eventually to be expressed in
terms of the temperatures of the nodal points, the element must possess three nodal
points. These nodal points are located so that they define the element geometry, i.e.
we obtain the triangular element shown in Figure 7.2l. X
The interpolation given by (7.80) is illustrated in Figure 7.22. As this interpolation
is linear in x and y the element is termed the linear triangular element. It also follows
Figure 7.22 Variation of temperature over linear triangular element
that contour curves within the element, i.e. curves for which T = constant, are straight
lines. values at the nodal points, i.e.
In order to determine the element shape functions corresponding to (7.80) we
again use the C-matrix method. Therefore, let us write {7.80) as T = Nf1; + N j 1j + Nt1k (7.83)
T= NIX (7.81 )
where, as before, NH x, y) is the element shape function at nodal point i and 7;, 7],
where 1k are the temperatures at the nodal points. Therefore, inserting the coordinates for
the nodal points, we obtain from (7.80) the following system of equations in the same
manner as for one-dimensional elements:
(7.82)
a• = CIX (7.84)
where
Just as for one-dimensional elements, our objective is to reformulate (7.80) so that

··~[n c~[: ']

it clearly appears that the temperature within the element is interpolated between its X;

xi YJ (7.85)
xk Yk
y
Moreover, determination of IX by means of (7.84) and insertion into (7.81 ) yields

(7.86)

where
N• = NC - 1 (7.87)
and

N • =[Nf Nj Nk] (7.88)

~---------------------------- x

Figure 7.21 Simplest possible two-dimensional element It can be observed that {7.86) is in accordance with {7.83).
120 Introduction to the finite element method Approximating functions- scalar problems 121

To determine the inverse matrix C - 1 , the determinant of C is calculated first. We observe that the element shape functions are linear functions of x and y in
From (2.40) and choosingj = 1 we find accordance with (7.80) and their variation over the element is shown in Figure 7.23.
By inspection it is easily shown that (7.57) again holds, i.e.
(7.89)
Moreover, as the coordinates of the nodal points are known, trivial geometrical 1 at nodal point i
arguments show that the area A of the triangle in Figure 7.21 is given by { 0 at all other nodal points (7.93)
Ni =
2A = det C (7.90)
Indeed this is a general property for all element functions and it follows directly from
One point needs some attention. In the matrix a• given by (7.85) we have listed t~e
the form (7.83) in a similar manner as in the discussion related to (7.57).
nodal points in the sequence i, j and k. Referring to Figure 7.21, it appears that th1s
Apart from the fundamental property given by (7.93) we also observe from Figure
sequence corresponds to a counter-clockwise direction. If a clockwise direction was
7.23 that, for instance, the element shape function Nk(x, y) belonging to the nodal
chosen, as given by the sequence i, k and j, for instance, this implies that the last two
point k is zero along the element boundary given by i, j. To prove this formally we
rows of C, as given by (7.85}, are interchanged. According to property (5) ~or
recall that Nk(xi, Yi) = N:(x1, y1) = 0, and as Nk( x, y) varies linearly it follows that
determinants (see page 18) this would change the sign of the . determinant, I.e.
det C = - 2A. For convenience, we want (7.90) to apply always, 1.e. we adopt the
N: = 0 along the element boundary i, j.
Let us return to the interpolation formula (7.83). To appreciate fully how it
convention that the nodal points should always be taken in the counter-clockwise
works, the illustration shown in Figure 7.24 is essential.
direction, as expressed through the sequence i, j and k. However, we may equally
In the weak formulation of the two-dimensional heat flow problem (cf. (7. 1)),
well take the sequences j, k, i or k, i, j which are obtained from the sequence i, j, k
the temperature gradient VTenters the expression. Therefore, from (7.86) we derive
by a so-called cyclic permutation. The reason is that each cyclic permutation d?es
not change det C. To see this, consider the sequence j, k, i. From (7.85) the followmg
aT] ox a•]
OX = [aN·
is implied: firstly that the first and second rows are interchanged, which ch~nges t~e
sign of det C; secondly that the second and third rows are interchanged, which agam VT =
changes the sign of det C so that it returns to its original value.
We conclude that for any counter-clockwise direction i, j and k where nodal
[ooyT oN•oy
a•

point i is arbitrary, the relation (7.90) holds.

The inverse matrix c- 1 can be established by means of (2.46) and some algebra which can be written as
will show that
(7.94)
XkYi - XiYk XiYJ - XJYi]
(7.91) where VN• is defined according to
Yk - Yi Yi-YJ
xi-xk x 1 -xi

VN· =
oN•]
OX
The element shape functions are obtained from (7.87) and (7.88), which with (7.95)
(7.82) and (7.91) yield oN·
[ oy
1 Let us define the matrix s• by
Ni = - [x 1yk- XkYJ + (yl- Yk)x + (xk- xJ)y]
2A
1
Nj = - - [xkyi- xiyk + (yk- y;)x +(xi- xk)y]
2A
(7.92)
s• = VN• =
oN•]
ox (7.96)
Nf. =-
1
[xiyJ- x 1yi + (y,- y)x + (x1 - xi)y]
oN•
[ oy
2A
122 Introduction to the finite element method Approxima ting functions - scalar p roblems 123

T·I

+ +
k

T·
J
I
I
I
I
I
I
I
Njl x. y) I
I
I

k
k~J
X Figure 7.24 Approximation of temperature in triangular element

which implies that (7.94) takes the form

VT = B•a• (7.97)

This expression for the temperature gradient is analogous to the formulations given
for one-dimensional elements (cf. (7.34) and (7.61)). The matrix B• relates the
temperature gradient in the element to the temperatures at the nodal points of the
element.
With N• given by (7.88) it follows that

X
oN't oN~
_J

B• = ox ox (7.98)
[ oN't oN•
_J

Figure 7.23 Element shape functions for linear triangular element oy oy

124 Introduction to the finite element method Approximating functions- scalar problems 125
y

Figure 7.26 Global shape function N;(x, y) for nodal point i for linear triangular elements
L---------------------------X
Figure 7.25 Two neighbouring triangular elements
over the entire body can be written as

The element shape functions are given by (7.92), i.e. T=Na (7.102)

The column matrix a contains the temperatures at all nodal points of the body, i.e.
(7.99)
Tl
T2
which shows that Be is a constant matrix. Therefore, in accordance with the linear a= (7.103)
expression (7.80), the temperature gradient VT is constant within the element.
Having established the elementwise approximation, our attention is now directed
T,
towards the temperature approximation over the entire body. For this purpose we
consider two neighbouring triangular elements (Figure 7.25). The element shape where n is the total number of nodal points in the body. The global shape function
functions Nt and Nf both relate to nodal point k, but Nt refers to element 1 and matrix N is then given by
Nf refers to element 2. From (7.93) it follows that Nt(xk> yk) = Nf(xk, Yk) = 1 and N=[N 1 N 2 000
N.] (7.104)
Nt(xi, Yi) = Nf(xi, Yi) = 0, and as both Nt are Nf are linear functions, we conclude
that N t = Nf holds along the common element boundary j, k. The temperature gradient within the body becomes
Considering a specific nodal point, let us construct a global shape function related
to that point in the same manner as for one-dimensional elements ( cf. (7.65) and I VT=Ba (7.105)
(7.66)). Therefore, for a given position (x, y), we may identify the associated element
IX which contains this position, i.e. where B is defined by

position ( x, y) =>identification of associated element IX (7.100) [aN, aN2

- --
ax
000

aN.]
ax
B = VN = ax (7.106)
For a given nodal point i the global shape function N; is then defined from the aN 1 aN 2 aN.
--- 000

element shape functions Nf according to ay ay ay

Nf(x, y) if element IX contains global nodal point i The approximation of the temperature given by (7.80) clearly fulfils the
N;(x, y) = { . (7.101) completeness requirements. Let us now check whether the simple triangular element
0 otherw1se considered is conforming, i.e. whether the approximate temperature distribution
varies continuously over common element boundaries. For this purpose consider
With these definitions the variation of a global shape function is shown in Figure 7.26. the common element boundary given by j, k in Figure 7.25. As the temperatures
From the global shape functions defined above, the temperature approximation at nodal points j and k are the same for the two elements and as the temperature
126 Introduction to the finite element method Approximating functions- scalar problems 127

varies linearly between nodal points j and k, it foUows that it varies in a continuous between the temperatures at the nodal points using the element shape functions, i.e.
manner over the common element boundary j , k. This means that the element is
conforming. T = N•a• (7. 108)
The linear triangular element dealt with so far is a very simple element. It was
suggested by Turner et al. ( 1956), and even though in many respects it marked the where
beginning of the FE method, it is still widely used. It is essential that the element Tl
can take the form of an arbitrary triangle, which implies that the geometry of any
irregular two-dimensional body can be approximated as closely as required. This T2
N• = [N~ N~ Nj N~]; a• = (7.109 )
ability to model arbitrary geometries is one of the essential advantages of the FE T3
method.
T4
and where the element shape functions fulfil the usual requirement given by
7 .3.2 Four-node rectangular element - the Lagrange element (7.93). This form may be obtained by expressing the parameters a 1 ... a 4 in terms of
the temperatures at the nodal points in the traditional manner using the C-matrix
Let us now consider an element which contains four nodal points, i.e. nodes. These method (cf. (7.84)). However, another method is possible using Lagrange's
nodal points are chosen so that they define the geometry of the element and we are interpolation formula (7.78).
then led to the rectangular element shown in Figure 7.27. For reasons that will be We first observe that in the x-direction, i.e. when y = constant, (7.107) implies
revealed later, the sides of the element are parallel to the coordinate axes. As we have that T varies linearly with x. Likewise, in the y-direction, i.e. when x = constant,
four nodal points, (7.2) and Pascal's triangle result in the approximation (7.107) shows that T varies linearly with y. Let us consider next, for instance, nodal
point 1 of Figure 7.27. In the x-direction we adopt Lagrange's interpolation formula
(7.107)
for n = 2 given by
The element defined by Figure 7.27 and (7.107) is occasionally referred to as the
Melosh element (Melosh, 1963), and it involves the parasitic term xy. Instead of this ll(x) = x- x 2
quadratic term xy, one may in principle choose the term x 2 or y 2 , but the choice xy xl- x2
is in general preferable because it implies that the dependence on x and y is similar, which fulfils /l{x 1 ) = 1 and ll{x 2 ) = 0. In they-direction we choose
i.e. the approximation is of the same type in the x- and y-directions. Despite the
presence of the quadratic term xy, it is also observed that (7.107) for y =constant l~(y) = Y- Y4
implies a linear variation with x and for x = constant a linear variation with y. For Y1- Y4
this reason, (7. 107) is often referred to as a bilinear expression.
i.e. Jl(yd = 1 and lf(y 4 ) = 0. The element shape function N~ is then given as
As before, we want to reformulate (7.107) so that the temperature is interpolated
x- x2 y- Y4 1
N~(x, y) = Jl(x)n(y) = = - ( x - x2)(y- y4)
X1 - X2 Y1 - Y4 4ab

y y

~-------------------------------- x
~-----------------------x

Figure 7.27 Four-node rectangular element Figure 7.28 Element shape function N: for four-node rectangular element
128 Introduction to th e finite element m e thod Approxim ating functions - scalar problems 129

where 2a = x 2 - x 1 = x 3 - x 4 and 2b = y 4 - y 1 = y 3 - y 2 (cf. Figure 7.27). The ·where

reader may verify that NHx 1 , y1) = 1 and NHx 2 , y2 ) = NHx 3 , y 3 ) = Nl(x4, Y4) = 0
in accordance with the general requirement given by (7.93). Using the same approach
for the other nodal points, we find that
aN·] [aN~
B· = VN· = ax = OX
aN"z
- - -aN;
--
ax ax ax
aN~]
(7.112 )
[aN·
ay
-
oN1
ay
--
oN"z aN3 -oN~-
-- - -
oy ay oy
e 1
N1 =- (x- Xz)(y- Y4)
T he element shape functions are given by (7.1 10), i.e.
4ab
1
N"z = - - - (x - x 1 )(y - Y3)
(7.113 )
4ab (7.110)

With the global shape functions defined in the usual manner ( cf. (7.1 00) and
(7. 101)) we obtain similarly to (7.102)- (7.1 06)

T=Na (7.114)

By inserting (7.110) in (7.108) it can easily be checked that the form given by (7.107) and
is obtained. Since the element shape functions may be derived using Lagrange's
interpolation formula, the fo ur-node rectangular element is an example of a Lagrange VT = Ba (7.115)
element.
The variation of one element shape function over the element is illustrated in where
Figure 7.28. The appearance of the other element shape functions is quite similar. oN
Whereas the element shape functions vary linearly for lines parallel to the coordinate
axes, the presence of the term xy in (7.107) implies that they vary non-linearly along B = VN =
[aN,
ox ax
2 ... ON,]
ox
(7.116)
all other lines. We observe that an element shape function related to a specific nodal oN 1 oN2 aN.
--
point is zero along element boundaries which do not contain the nodal point in ay ay ay
question. This property was also found for the triangular linear element and in order
The approximation (7.107) fulfils the com pleteness requirement (cf. (7.2)). To
to prove this property formally for the four-node element, consider, for instance, the
check that the element is conforming, i.e. that the temperature varies continuously
element shape function N~ shown in Figure 7.28 along the element boundary 2,3.
across common element boundaries, consider the two elements shown in F igure 7.29.
As x = constant along element boundary 2,3, N~ varies linearly with y along this
Along the common boundary 2,5 we have x = constant, i.e. according to ( 7. 107)
boundary. Moreover, as N~(x 2 , y2 ) = N~(x 3 , y 3 ) = 0, we conclude that N~ = 0 along
element boundary 2,3. The same type of argument implies that if a nodal point is
shared by two neighbouring elements, then the two element shape functions belonging y
to that nodal point are identical along the common element boundar y.
4 5 6
As before, the temperature gradient is of interest and similar to (7.94)- (7.98 ) we

I l l
obtain from (7.108)

VT = VN•a• CD cv
which can be written as 2 3
X
VT = B•a• (7.111)
Figure 7.29 Conforming elements
130 Introduction to the bnite element method Approximating functions- scalar problems 131

y be shown that the elements in Figure 7.31 are conforming, i.e. continuity across
common element boundaries exists.
The restriction that the four-node element must be rectangular with its element
boundaries parallel to the coordinate axes has been emphasized. However, we shall
see in Chapter 19 that this restriction can be removed by means of the so-called
isoparametric formulation of finite elements.
2 3
7 .3.3 More complicated rectangular and triangular
L------------------------------------ x
elements - serendipity elements
Figure 7.30 Non-conforming elements
It is obvious that the manner in which Lagrange's interpolation formula was used
y to construct the element shape functions for the four-node rectangular element can
be generalized. Without going into detail, consider the rectangular element shown
in Figure 7.32. The specific feature of this six-node element is that it contains three
nodes in the x-direction and only two nodal points in the y-direction. Using Pascal's
triangle the temperature approximation becomes
T = o: 1 + o:2x + cx 3y + o:4x 2 + o: 5xy + o:6 x 2y (7.117)
which implies a quadratic variation in the x-direction and a linear variation in the
y-direction, in accordance with Figure 7.32. The element shape functions may be
Figure 7.31 Conforming elements constructed directly in a manner similar to that leading to (7.110), i.e. any element
shape function is obtained as a suitable multiplication of a second-order Lagrange
polynomial in x and a first-order Lagrange polynomial in y. The element of Figure
the temperature varies linearly withy for both elements. As the temperatures at nodal 7.32 is therefore another example of a Lagrange element and may be of interest for
points 2 and 5 are identical for the two elements we conclude that continuity exists problems where it is known, a priori, that the temperature variation is more
across the common element boundary; that is, the element is conforming. pronounced in the x-direction than in the y-direction.
However, it is important that this continuity is present only when the element Another Lagrange element that is used more often, and which exhibits
boundaries are parallel to the coordinate axes. To show this consider Figure 7.30. geometrical isotropy, is the rectangular nine-node element shown in Figure 7.33. As
The common element boundary 2,5 is given by the equation y = ax + b. Insertion we have a quadratic variation in the x - and y-directions and nine nodal points, use
of this expression into (7.107) implies that of Pascal's triangle yields the following approximation:
T = o: 1 + o: 3b + (o: 2 + o:3a + o:4b)x + o:4ax 2 T = o:I + cx2x + cx3y + o:4x2 + O:sXY + o:6y2 + a1x2y + O:sXY2 + o:9x2y2
i.e. T varies quadratically with x along boundary 2,5. This quadratic function is (7.118)
defined uniquely by three temperatures, but as only the two temperatures T2 and T5 y
are known, the quadratic expression above is not uniquely determined. This means
that the temperature variation along the common boundary in element 1 differs from
that of element 2. Therefore, continuity across common element boundaries is
violated implying that the element is non-conforming. In general, we would like to
avoid this situation, which means that the four-node element should be used in a
form where the element boundaries are parallel to the coordinate axes.
It appears that the four-node rectangular element in itself is unable to model
arbitrary geometries of the body. However, if it is combined with the simple triangular
element, arbitrary geometries can be modelled, and an example of such a combination
is shown in Figure 7.31. Using arguments similar to the above discussion it can easily Figure 7.32 Six-node rectangular Lagrange element
132 Introduction to the finite element method Approximating functions- scalar problems 133
y y

L------ -- - - - - - x
Figure 7.33 Nine-node rectangular Lagrange element Figure 7.35 Six-node triangular element

y
0

Figure 7.34 Conforming elements

1\ 0

L------------------x
Figure 7.36 Conforming elements
which is symmetric in x and y. It appears that the last three terms a re parasitic.
Moreover, as aU linear and quadratic terms are present, this nine-node element is Using arguments similar to those given previously, it can easily be shown that
able to reflect an arbitrary linear variation of the temperature gradient. The element any element mesh consisting of six-node triangular elements is conforming. As the
shape functions can be obtained as a suitable multiplication of Lagrange polynomials six-node element may form any triangle, we conclude tha t any two-dimensional
of second order in the x- and y-direction or using the C-matrix method. As the geometry can be approximated as closely as requested. As shown in Figure 7.36, the
temperature varies parabobcally along any element boundary, a combination of six-node triangle can also be used in combination with the rectangular nine-node
rectangular nine-node elements is conforming. However, analogous to the discussion element to model bodies with skew boundaries and this combination is easily shown
of Figure 7.30, this conforming behaviour requires that the element boundaries are to be conforming. The six-node triangle is treated in detail by, for instance,
parallel with the coordinate axes. This requirement can be relaxed by use of the Zienk:iewicz and Taylor {1989) and Gallagher {1975).
isoparametric FE formulation presented in Chapter 19. A conforming combination At this point, it is no surprise that a variety of elements exists and can be
of the three-node triangle and the four-, six- and nine-node rectangular Lagrange suggested, and we shall conclude the discussion of two-dimensional elements by
elements is shown in Figure 7.34. Further information on the nine-node Lagrange considering the often used eight-node rectangular element shown in Figure 7.37.
element may be found, for instance, in Becker et al. ( 1981) and Hughes ( 1987). The eight-node element looks a little like that of Figure 7.33, but it is not a
It is also possible to obtain triangular elements of higher order than the simple Lagrange element because we only have two nodal points along the line given by
three-node element treated previously. An example is the six-node triangle shown in nodal points 5 and 7. Therefore, Lagrange polynomials cannot be used directly to
Figure 7.35. Referring to Pascal's triangle, we can write the approximation as establish the element shape functions. Nevertheless, it is possible to construct these
T = tX 1 + tX 2 x + o: 3 y + 0: 4 X 2 + tX 5 xy + tX 6 yl (7.119) element shape functions directly. To facilitate this process, we make a coordinate
transformation from the xy-system to the ~17-system (cf. Figure 7.37). The
which forms a complete polynomial suggesting an efficient element. Moreover, any ~17-coordinate system is called a local coordinate system and it is devised so that
linearly varying temperature gradient can be modelled by (7.119). 1~1 = 1 and 1111 = 1 hold along the element boundaries. Clearly, it is easier to establish
134 Introduction to the finite element method Approximating functions - scalar problems 135
y
7 3
1- 1) = 0
''
1) = 1 ' '8 6
7 I 3
~=-1----.... -~=1

8 6

5
Figure 7.39 Construction of element shape function N~ for corner node

L-----------------------------------x
Next let us establish the element shape function for a typical corner node like
Figure 7.37 Eight-node serendipity element nodal point 1. Referring to Figure 7.39, we require that N~ is zero along the sides
2,6,3 and 4,7,3. Moreover, N~ is also required to be zero at nodal points 5 and 8. It
7 3 turns out to be advantageous to assume that N! is zero along the line defined by
1-1)=0 points 5 and 8. These observations imply that Nl = -±(1 -
~)( 1 - '7 )( 1 + ~ + '7 ),
where the factor -± is determined such that NH -1, -1) = 1. Following the same
8 6 procedure for the other corner nodes, we obtain

N[ = -±(1 - 0(1- 17)(1 + ~ + '7)

5 N~ = -±(1 + ~)(1 - t7)(1- ~ + '7)
( 7.121 )
N; = -±(1 + ~)( 1 + t~H1- ~- t7)
N~ = -±(1 - ~)(1 + t~H1 + ~- f/ )
Figure 7.38 Construction of element shape function N5 for mid-side node
Even though we have presented a systematic approach fo r the establishment of the
element shape functions, these functions were originally formulated by inspection
(Ergatoudis et al., 1968). The eight-node element is therefore called a serendipity
the element shape functions in the local ~17-coordinates, and from these results the element after the famous princes of Serendip noted for their chance discoveries
element shape functions may be expressed in terms of xy-coordinates, if requested. (Horace Walpole, 1754 ). The element shape functions for a typical mid-side and
Let us first consider a typical mid-size node like nodal point 5 in Figure 7.38. It • corner node are illustrated in Figure 7.40.
is required that the element shape function N5 is zero along the sides 2,6,3 and As before, we have that the temperature Tis related to the temperature at the
4,7,3 and 1,8,4. From Figure 7.38 we conclude that N5 is given by N5 = nodal points a• through T = N•ae, and using (7.120 ) and (7.121) it follows that
!(1 - ~)( 1 + ~)( 1 - 17 ), where the factor ! is determined such that N5(0, - 1) = 1.
By similar arguments for the other mid-side nodes we obtain (7.122 )
We conclude that the eight-node serendipity element can model any linear variation
N5 =to - 2
~ )(1- 11) of the temperature gradient and that (7.122) is symmetric in x and y. The last two
+ ~)(1- '7 2 )
N(, = }(1 terms of (7.122) are parasitic of third order and a comparison with the nine-node
(7.120) Lagrange element (Figure 7.33 and (7.118)), shows tha t the internal node of the
N; = t(l- ~ )(1 + 17)
2
Lagrange element contributes a parasitic term of fourth order. This suggests that for
N8 = t(l- ~)(1- '7 2 ) the same number of nodal points (i.e. degrees of freedom) in a body, the eight-node
serendipity element is slightly more efficient than the nine-node Lagrange element.
136 Introduction to the finite element method Approximating functions- scalar probl ems 137
a) z
z
a) b)
k
4

b)

X
X

Figure 7.41 (a) Four-node tetrahedral; (b) ten-node tetrahedral

Figure 7.40 Element shape functions for eight-node serendipity element: (a) mid-side node;
(b) corner node

As before, the eight-node serendipity element must have element sides parallel to the
coordinate axes in order to be conforming. This obstacle can be removed by using
the isoparametric formulation presented later. 8 nodes 20 nodes
We emphasize that instead of the derivation prescribed above, it is always
Figure 7.42 Eight- and twenty-node prism element
possible to establish the element shape functions by means of (7.122) in combination
with the C-matrix method.
shown in Figure 7.41 (b). The approximation becomes

7. 4 Three-dimensional elements
(7.124)
Three-dimensional or solid elements can be devised directly from the exposition above.
A detailed discussion of three-dimensional elements is given in Hughes (1987), which is a complete polynomial in accordance with the approximation for the
Gallagher ( 1975) and Zienkiewicz and Taylor (1989), but as no new conceptual six-node triangle (cf. (7.119)).
matters are involved, we shall confine ourselves here to some rudimentary Three-dimensional elements in the form of prisms or bricks with element sides
facts. parallel to the coordinate axes are easily derived, for instance from the four-node
The simplest three-dimensional element, the tetrahedral shown in Figure element of Figure 7.27 and the eight-node element of Figure 7.37. These solid elements
7.41(a), has four nodal points corresponding to the approximation are illustrated in Figure 7.42 and the temperature approximation for the eight-node
prism becomes
(7.123)
The generalization of the six-node triangle results in the ten-node tetrahedral as (7.125)
138 Introduction to the finite element method Approxima ting functions- scalar problems 139

whereas the approximation for the twenty-node isoparametric element is given by In addition, we found that an element shape function related to a specific nodal point
is zero along element boundaries which do not contain the nodal point in question.
T = a 1 + a 2 x + a 3 y + a 4 z + a 5 x + a6 y + a 7 z + a 8 xy + a 9 xz + a 10 yz
2 2 2
Let us now prove another general property for element shape functions. All the
+ a 11 x2 y + a 12 xy2 + a 13 x2 z + a 14 xz 2 + a 15 y2 z + a 16 yz 2 + a 17 xyz elements considered fulfil the completeness requirement, i.e. the approximation is
always of the form given by (7.2). By putting all terms except a 1 equal to zero, it
(7.126)
appears that all the elements considered are able to describe a constant temperature
If x, y or z is assumed to be constant, i.e. if one of the sides of the bricks is considered, within the element. If this temperature is called Yc we obtain from (7.127) and (7.128)
it follows that (7.125) and (7.126) reduce to (7.107) and (7.122), respectively. Using that
the isoparametric formulation presented later, these three-dimensional elements may
also be used in a conforming manner even when the element sides are not parallel
1'c = N~ 1'c + N21'c + .. · + N~ 1'c
with the coordinate axes. t.e.

1 k1~1 1
(7.130)

7.5 Summary By inspection we may convince ourselves that this relation holds true for the element
shape functions derived previously.
We have seen that the elementwise approximation technique offers a systematic and From (7.127) the temperature gradient within the element is given by
unified approach of obtaining an approximation for the unknown function - the
temperature - over the entire body. The possibility for treating bodies with arbitrary I VT = B•a• (7.131 )
geometries is a hallmark of the FE approximation technique. In addition, the freedom
to choose elements with different sizes as well as different types greatly enhances the For one-dimensional elements we have
possibilities offered by this approximation technique.
Irrespective of the particular element chosen, the approximation of the unknown dT
VT=-· B• = [ dN~ dN2
-- dN~J (7. 132)
function - the temperature - is carried out as an interpolation between the nodal dx ' dx dx dx
points of the element. This is expressed through
Two-dimensional elements imply that

~ [~:J
T = N•a• (7.127) oN~

B" =
[ON"
ox
1
--
ox
... ON~]
OX
where N• is the element shape function matrix and a• contains the temperature at VT (7.1 33)
oT ' oN~ oNi oN"n,
the nodal points. That is, --
oy oy oy oy
whereas three-dimensional elements yield
N• = [ N~ N2 . . . N~J; a• = (7.128)
oT oN~ oN2 oN~
--
ox ox ox ox
T,.,
oT oN~ oN2 oN~
where n. denotes the number of nodal points in the element. The element shape VT = B• = -- (7.134)
functions, N~, depend on the coordinates. We recall the following fundamental
oy oy oy oy
property for the element shape functions: oT oN1 oNJ. oN~
oz oz oz oz
1 at nodal point i
(7.129) To formulate the approximation of the temperature over the entire body we
Ni = { 0 at all other nodal points
introduce the global shape functions similarly to (7. 100) and (7.101 ). That is, first
140 Introduction to the finite element method Approxima ting functions - scalar problems 141

we observe that any position (x, y, z) identifies the associated element rx which Two-dimensional elements imply
contains this position. Therefore

VT~ [~:J
[aN, oN2 ...
position (x, y, z) =identification of associated element rx (7.135) B= OX ax aN.]
ox
(7.142)
aT ' aNi 8N2 aN.
For a given global nodal point i the global shape function Ni is then defined from
--
oy oy ay ay
the element shape functions Nf according to
whereas three-dimensional elements results in

_ {Nj(x, y, z) if element a contains global nodal point i aT aNt oN 2 oNn

Ni ( x,y, z ) - . -
0 otherwise ax ax ax ax
aT oNl oN2 aN.
(7.136) VT= B= - - - - (7.143)
ay ' ay ay ay
It is of fundamental importance that these definitions imply that a global shape aT oN 1 oN 2 aN.
function related to a specific nodal point differs from zero only in those elements -
which contain the nodal point in question.
oz az az az
Using these preliminary remarks the approximation of the temperature over the A variety of elements has been discussed and it seems natural to ask whether it
entire body is given by is possible to nominate the best element. Unfortunately, this is not possible in general
because the 'best' element often differs from problem to problem. However, we may
T=Na (7.137) state that, as the order of an element increases, the total number of unknowns in a
problem can in general be reduced for a given accuracy of the approximation. This
where N is the global shape function matrix and a contains the temperatures at the is especially true for three-dimensional elements. Use of higher-order elements will
nodal points of the entire body. That is, also reduce the amount of data preparation for the FE program, but this advantage
may be offset by the more involved numerical manipulations. In practice, it is rare
to use elements of higher order than those discussed previously.
Apart from these general remarks, we note that in regions where the gradient
(7.138) varies rapidly, more elements are clearly needed in order to capture this variation
than in regions where the gradient varies slowly. Finally, we recall that in order to
obtain a conforming behaviour of two-dimensional rectangular and three-dimensional
where n denotes the number of nodal points for the entire body. As expression (7.137) brick elements, the element sides must be parallel with the coordinate axes. This
is able to reflect a constant temperature over the entire body in an exact manner, essential drawback can be removed by using the isoparametric formulation presented
we derive in a way similar to (7.130) that in Chapter 19.

I ,t,N, ~ 1 I (7.139)

From (7.137) the temperature gradient is given by

VT= Ba (7.140)

For one-dimensional elements we have

VT = dT. B = [dN1 dN 2 dNn] (7.141)

dx' dx dx dx
 Choice of weight function 143

8 method is obtained by the Galerkin method (Galerkin, 1915) which is also used in
other formulations than the FE approach. More generaJiy, the Galerkin method is
Choice of weight function an example of a so-called weighted residual method, and in order to substantiate our
choice of the Galerkio method as an integrated part of the FE formulation, we will
weighted residual methods discuss different weighted residual methods in this chapter.
General theories for weighted residual methods may be found in Crandall ( 1956),
Finlayson and Scriven ( 1966), Finlayson ( 1972) and Zienkiewicz and Taylor (1989).
However, our intention is not to provide the reader with comprehensive knowledge
of different weighted residual methods, but simply to obtain sufficient motivation for
our choice of weight function in terms of the Galerkin method. With this limitation
in mind we consider the following one-dimensional differential equation applicable
in the range a~ x ~ b:

Let us return to our model problem, namely that of heat flow. The strong form of Lu + g = 0; a ~x ~b (8.2)
the problem was first established and, based on tills result, we derived the
corresponding weak formulation. The advantages of the weak form as compared where u(x) is the unknown function and g(x) is a known function ; L denotes a
with the strong form have already been discussed and considering, for example, differential operator. When L is chosen, it specifies the actual form of the differential
two-dimensional heat flow, the weak form was given by (6.46), whlch reads equation given by (8.2). As an example, L may be given by L = d/ dx, which implies
that (8.2) takes the form du f dx + g = 0. As another example, for L = d 2 / dx 2 + 1,
(8.2) implies the differential equation d 2 uj dx 2 + u + g = 0. It appears that by using
(Vv)TtDVT dA = - f vhtd£!1 - f vqntd£f+f vQtdA (8.1)
fA • ~ A
the format (8.2), it is possible to write quite general one-dimensional differential
equations in a compact fashion. In order to simplify the discussion, the boundary
The FE method is an approximate method for solving arbitrary differential equations. conditions for (8.2) are assumed to be given by
Therefore, an approximation is required for the unknown function, which in the
present case is the temperature T. In the previous chapter, a detailed di_scussion w~s (8.3)
devoted to the specific type of approximation for the temperature that IS adopted 10
where ua and ub are known quantities. The specification of two boundary conditions
the FE method. This approximation was found to be characterized as an elementwise implies that the differential equation ( 8.2) is of second order.
approximation over the entire body. Referring to (8.1), it appears that the only topic It is characteristic that the differential equation (8.2) with the boundary
which remains to be discussed is the choice of the weight function v. We recall that conditions (8.3) can only be solved analytically in terms of an exact closed-form
in the weak formulation this weight function is an arbitrary function. At this point solution for certain simple expressions for the differential operator L and the function
we can therefore summarize the basic steps in the FE formulation as follows: g. What we are looking for are methods which enable us to solve (8.2) subject to
(8.3) for arbitrary expressions for L and g. This gain in generality is paid for in the
sense that our solution takes the form of an approximate solution.
Basic steps in the FE formulation For this purpose, we first make a slight reformulation of (8.2). We may multiply
1. Establish the strong formulation of the problem. (8.2) by an arbitrary function v(x), the weight function, to obtain
2. Obtain the weak form of the problem.
v(Lu+g)=O
3. Make an elementwise approximation over the entire body of the
unknown function, i.e. the temperature T.

r
and we may even integrate this expression over the region of interest, i.e.
4. Choose the weight function v.

v(Lu +g) dx =0 (8.4)

The previous chapters have been devoted to steps 1- 3 and in the present chapter
we shall establish the choice made for the weight function v, i.e. step 4. It is obvious that (8.2) implies (8.4) and as v denotes any function, it is easily shown
It turns out that a particularly suitable choice of the weight function in the FE that (8.4) implies (8.2), i.e. the two forms are equivalent. To show that (8.4) implies

142
144 Introduction to the finite element method Choice of weight function 145

(8.2), choose the arbitrary weight function vas v = Lu + g, i.e. (8.4) becomes surprising that u•PP will not, in general, satisfy the equation exactly. That is, we obtain
Lu•PP + g = e (8.9)
r(Lu+g)2 dx=0
where e(x) is a measure for the error. This measure is termed the residual. With (8.9),
(8.8) can be written as
which can only be true if (8.2) is fulfilled. We have thus proved that (8.2) and (8.4)
are equivalent. It is apparent that this discussion has much in common with that of
the weak formulation for one-dimensional problems (see Chapter 4). However, it is (8.10)
important to realize that the formulation (8.4) is not a weak formulation. In order
to achieve a weak formulation an integration by parts is necessary, whereby the order
of differentiation of the unknown function is decreased at the expense of the weight Since the residual e depends through (8.9) and (8.6) on the unknown parameters
function being differentiated (cf. for instance ( 4.36)). given by a, (8.10) is an expression which depends on the parameter a 1 ... a•. The
With these remarks in mind, we return to the formulation given by (8.4). As we objective is to choose the weight function v so that these parameters can be
seek a numerical solution, an approximation must be chosen for the unknown determined, which, in turn, implies that the approximate solution (8.6) is known.
function u. Quite generally, we make the following assumption, which is assumed to Equation (8.10) may be interpreted so that the residual, e(x), is given a certain
fulfil the boundary conditions (8.3): weight, v(x), and the integral of this weighted residual, v(x)e(x), over the region of
interest is required to be zero. The various weighted residual methods differ in how
(8.5) the weight function v is chosen. Obviously, the specific choice of the weight function
Here a 1 ... a. are unknown parameters and t/1 1 t/1. are functions of x, which we influences the values of the parameters a 1 ••. a., which are to be determined. Another
specify in advance. These functions are called trial or basis functions. Once the feature also related to (8.10) is that of orthogonality. The orthogonality condition for
parameters a 1 ••• a. are known, our approximate solution is given by (8.5), i.e. the two column vectors a and b is expressed through arb= 0. For functions, one can
task is to determine the parameters a 1 ... a• . The trial functions may be chosen also speak of orthogonality and the function v(x) is said to be orthogonal to the
arbitrarily, but it is obvious that some knowledge of the physical problem facilitates function e(x) in the interval a~ x ~ b if (8.10) holds.
the choice of suitable trial functions. With reference to Chapter 7, we may even take Equation (8.10) serves as a vehicle to determine the unknowns a 1 ... a., i.e. a.
the approximation (8.5) as an elementwise approximation. In that case al ... a. To see how this can be accomplished, we write the arbitrary weight function quite
would be the values of u at the nodal points and the trial fun~s would become generally as
the global shape functions. However, (8.5) may be taken as any approximation and v = Vlct + V2 c2 + ... + V,c. (8.11)
this is the reason why t/1 1 ... t/1" are termed trial functions.
The approximation (8.5) may be written as where Vl . . . V, are known functions of x, which we specify in advance, and c1 . . . c.
are certain parameters. We note that the number of terms in (8.11) is the same as

I u•PP = rjla (8.6)

the number of terms in (8.5). With the definitions

where
v~cv, v, ... V.J; c{:J (8.12)

(8.7)
(8.11) can be written as

v= Vc (8.13)

r
We may substitute u in (8.4) by u•PP and require that
As the weight function is arbitrary and as V is known, we conclude that the
v(Luapp +g) dx = 0 (8.8) parameters c 1 ... c., i.e. c, are arbitrary. Moreover, as v is one number and as the
transpose of a number is equal to the number itself, i.e. v = vr, we may write (8.13) as
Likewise in the differential equation (8.2), u may be substituted by tfPP and it is not (8.14)
146 Introduction to the finite element method Choice of weight function 147

Inserting (8.14) into (8.10) and observing that cT does not depend on the coordinate showing that L(t/1) is a matrix with dimension 1 x n. Use of (8.17) in (8.1 5) implies

r
x, we obtain that
(8.19)
cT VTedx =0
As this expression should hold for arbitrary cT-matrices, it is concluded that

(8.15)
K= r yT L("') dx; f =- r
where a is considered to be independent of x. With the definitions

VT g dx

where both K and f are independent of a, (8.19) can be written as the system of linear
(8.20)

equations
It is important to realize that as the column matrix VT has the dimensions n x 1 (cf.
(8.12)), expression (8.15) comprises, in fact, n equations. That is, (8.15) may be written
as
I Ka = f I (8.21)

As yT has the dimension n x 1 and L{t/1) the dimension I x n (cf. (8.18)), the
J:v1 edx=0 expressions for K and f may be written as

J:v2edx=O r r VtL(t/lt) dx V,L{t/12) dx r

V1L (t/J.) dx J:v1gdx
' ?8.16)
f f f lb V2 g dx

r
V2L(t/1 1) dx V2 L(t/1 2) dx V2L{ t/ln) dx
K= ; f= -

v.edx =0

The derivation of(8.16) is based on (8.10) and (8.13) and uses the fact that the
weight function vis arbitrary, i.e. cis also arbitrary. Alternatively, in (8.10) one may
r r V,L ( t/1 1 ) dx V,L( t/1 2) dx r
V"L(t/1.) dx r V,g dx

(8.22)
directly choose n arbitrary weight functions v and first choose v = V1 , then v = V2
and so on. Clearly, the result of this approach is again given by (8.16). The approach showing that K is a square matrix with dimension n x n. That is, (8.21) consists of
based on (8.13) is preferred here, as it facilitates a short notation, especially in later n linear equations from which the n unknowns a 1 .. . a., i.e. a, can be determined.
FE formulations, but one should be aware of the equivalence of the two approaches When a has been obtained from (8.21), (8.6) provides the required approximate
both leading to (8.16). solution.
Returning to (8.16), the essential point is that the residual e depends on the n The procedure described above applies to all weighted residual methods, and a
unknowns a 1 ... a. (cf. (8.9) and (8.6)). Therefore (8.16) serves as a system of variety of different specific weighted residual methods is obtained depending on our
equations to determine these n unknowns. To see this more clearly, let us insert (8.6) choice of the weight function v, i.e. our choice for V (cf. (8.13 )). Some frequently used
into (8.9) to provide weighted residual methods will be described below.
e = L(t/la) +g
The a-matrix does not depend on the coordinate x, i.e. 8.1 Point collocation method
e= L(I/I)a+g (8.17) In this method, the weight function v is chosen based on Dirac's delta function
Following Chapter 2, differentiation of a matrix means that each component is c5(x- xJ This function is defined as
differentiated, i.e. we obtain from (8.7) that
L(t/1) = [L(t/1 1) L(t/1 2) . .. L(t/1.)] (8.18)
c5(x - X·)=
' 0
{oo ifx =X;
otherwise
(8.23)
148 Introduction to the finite element method Choice of w eigh t f unction 149

e(x )
00

--------~----------~~----------x

Figure 8.2 Point collocation method

Figure 8 .1 Dirac's delta function

and obtain from (8.22) in a manner analogous to (8.28)

(8.24) L(l/1 1(x 1)) L(l/fix 1)) L(l/Jn(xl))
L{l/1 1(x 2)) L(l/1 2 (x 2)) L(l/ln(Xz))
where x 1 is a given fixed value. Alternatively, due to (8.23), we may write (8.24) as K= (8.30)

f
xt
x-b(x- x;) dx =I (8.~)
I The differentiation indicated by the operator L , for instance L(l/1 1(x 2 ) ), should be
where x:+ and x:- denote x-values slightly larger than and smaller than xi, understood as follows : first the differentiation of the function l/J 1(x) is carried out
respectiv~ly. Using' these definitions, Dirac's delta function is iiiustrated in Figure 8.1. and then the specific x-value, x = x 2 , is inserted.
In the point collocation method, the weight function v in (8.13) is chosen such that
( 8.26)

The fixed points x 1 xn are chosen arbitrarily within the region a~ x ~ band are
00. 8.2 Subdomain collocation method
called collocation points. In order to illustrate the consequence of this choice of weight

r r
function, we evaluate (8.16), which becomes In the point collocation method n points were chosen. In the subdomain collocation
method, however, the region of interest is divided into n subregions. Each of these
Jlle{x) dx = b{x- x 1)e(x) dx = 0; i = I, ... , n (8.27) subregions is given by xi~ x ~ x 1+1 where 1 ~ i ~ nand both x 1 and x i+ 1 are located
in the region a ~ x ~ b. The matrix V ( cf. (8.12)), which defines the weight function,
As Dirac's delta function is zero unless x = x 1 (see (8.23)) we have is chosen such that

= {1 if X1 ~ X ~
fx~~(x - x )e(x) dx = e(xJ fx~~(x- xi) dx =
I
Xi + 1 . __
b !5(x- xJe(x) dx = e(x1) V, . z 1, ... , n (8.31 )
1 0 otherwtse
Q XJ X,
(8.28) as illustrated in Figure 8.3.

r
where (8.25) has been used. From (8.27) and (8.28) we obtain

V;e(x) dx = e(xJ = 0; i = 1, 00 0
' n

That is, the point collocation method forces the residual e(x) to be zero at the
(8.29)
This means that (8.16) becomes

Ia
b
V;e(x) dx =
I X/+1
x1
e(x) dx = 0; i = 1, ... , n
That is, the subdomain collocation method forces the average of the residual over
(8.32)

collocation points, as illustrated in Figure 8.2. each subdomain to be equal to zero, as illustrated in Figure 8.4.
In order to determine the specific form of the system of equations (8.21), we With (8.31) and using (8.22), the coefficient matrix K and t he right-hand side f
150 Introduction to the finite element method Choice of weight function 151

8.3 Least-squares method

Subdomain i
I· ·I Due to (8.17) we obser ve that the residual e depends on both x and the parameters
a 1 ... a., i.e.

b
(8.34)
a
L----r--+-----+-------~-----------;--~----- x
Xj Xn Xn•l In the least-squares method, a typical component of the matrix V given by (8.12) is
chosen as
Figure 8.3 Weight function in subdomain collocation method
V; = -; i
oe = 1, .. . , n (8.35)
oai
which implies that (8.16) takes the form
e l xl

Subdomain i \ f~ b
a oai
e dx = 0; i = 1, . .. , n (8.36)

In order to evaluate this choice, consider the following quantity I :

(8.37)

Figure 8.4 Subdomain collocation method

Since we integrate with respect to x, it follows that the quantity I depends on the
parameters a 1, ••• , a. only, i.e. I = I(a" .. . , a.). From (8.37) we conclude that
of ( 8.21) take the forms
(8.38)
x, L(l/1
f x,
1
) dx

K = f:,' L( 1/1 1) dx
An evaluation of (8.37) shows that I is the square of the error, i.e. the residual,
integrated over the region of interest. A comparison of(8.36) and (8.38) reveals that

a1
- = 0; i = 1, ... , n (8.39)
J:.•+• L(l/1.) dx oai

(8.33) i.e. our choice of weight function given by (8.35) implies that I is stationary. As the
residual e, and thus also the quantity I , can be made arbitrarily large, we conclude

f
Xl

g dx that the stationarity of I as expressed through (8.39) is a minimum. Consequently,

the choice of weight function given by (8.35) implies that the square of the error is
x~,

f
a minimum ; hence the terminology of the least-squares method.
g dx To determine specifically the coefficient matrix K and the right-hand side f of
f =- x,
the system of equations (8.21) when the least-squares method is adopted, we first
insert (8.1 7) into (8.35) to obtain

V; =L(I/IJ; i=1, ... , n (8.40)

152 Introduction to the finite element method Choice of weight function 153
That is, (8.22) yields (8.22) to provide

r L(l/Jt)L(l/1 1) dx rL(l/1 1)L{l/1 2) dx rL{l/1 1 )L(l/JJ d x r "'IL(l/11) dx r "'1L(l/12) dx rl/JtL(l/J.) dx

K = r L(l/12)L(l/IJ dx r L(l/1 2)L(l/12) d x r L(l/1 2)L(l/Jn) dx
K = r l/1 2 L(l/1 1 )dx r l/12L(l/12)dx f l/1 2 L(l/l.) dx

r L(l/ln)L(l/1.) dx
(8.41)
r l/J. L( l/1.) dx

r L(l/1 1 )g dx
(8.45 )

f= -
r L{l/J 2)g dx
f= -

r L(l/ln)g dx

which shows that the coefficient matrix K is symmetric.

8.5 Example
8.4 The Galerkin method
To illustrate the weight residual methods described above, we consider the differential
In the Galerkin method (1915) the components of the matrix V given by (8.12) are equation
chosen in accordance with
(8.46)
1 V; = "'i 1 i = •..... n (8.42)
In the notation of (8.2) it follows that
i.e. the components v; are equal to the trial functions l/1;- Loosely speaking we express d2
this by saying that L= - +
2
t· g=x (8.47)
dx '

weight functions = trial functions (8.43) The boundary conditions are given as

With (8.42), (8.16) yields u(O) = u(l) = 0 (8.48)

As easily proved by inspection, the exact solution to (8.46) and (8.48) is
r l/J;e dx = 0; i = 1, ... , n (8.44)
sin x
u= - - -x (8.49)
Referring to the discussion of (8.10), it follows that in the Galerkin method the trial sin 1
functions are orthogonal to the residual. To determine explicitly the coefficient matrix In order to obtain an expression for the approximation to u in the form of (8.5),
K and the right-hand side f of the system of equations (8.21), (8.42) is inserted into we may explore a number of possibilities. One possibility is given by the trigonometric
1S4 Introduction to the finite element method Choi ce of weight function ISS

series Next, consider the least-squares method, where K and fare given by (8.41) which
implies that
u•PP = b0 + (a1 sin ex+ b 1 cos ex)+ (a 2 sin 2ex + b 2 cos 2cx)
+ ··· +(a. sin ncx +b. cos ncx) (8.50)

This approximation is required to fulfil the boundary conditions (8.48) and it appears
K= [f L(t/t 1)L(t/t 1)dxJ= [11( 2
-n sin nx + sin nx) dx = 2 J G J
(1 - n2?

that u•PP(O) = 0 is fulfilled if

( 8.51)
f= -[fL(t/t1)g dx = -[f( -n2sin nx +sin nx)x dx J= -[~ (1 - n2)]
b0 = b 1 = b2 = · · · = b. = 0
Moreover, the condition u•PP( 1) = 0 is fulfilled if (8.60)

(8.52) and where it was used that

c=n
With (8.51) and (8.52), (8.50) reduces to
u•PP = a 1 sin nx + a 2 sin 2nx + · · · + a. sin nnx
\
(8.53) f ·2
sm x dx = -X2 - -41 sin 2x·, f x sin x dx = sin x - x cos x

As a very simple approximation, we take j ust one term in this series, i.e. From (8.21) it follows that
u•PP = a 1 sin nx (8.54)
1 2
2 2
- (1 - n ) a 1 = - -1 ( 1 - n 2) i.e. a 1 = ~ 0.0718 (8.61)
and the problem is now to determine the parameter a 1 . With reference to ( 8.6) we have 2 1t n(n 2 - 1)
(8.55)
Finally, from (8.45) the Galerkin method implies that
Consider first the point collocation method. From (8.47) and (8.30) we obtain

K = [L(t/t1 1))]
(x = [ -n 2 sin nx 1 +sin nx 1] (8.56)
K= [f t/t 1L(t/t1)dx = [fsin nx( -n2 sin nx +sin nx) dx J= G(1 - n2) J
f = - [g(x 1 )] = - [x 1 ]

Choose the collocation point, for instance, as the midpoint of the interval, i.e. x 1 = t. f= -[f t/t1g J -[1dx =
1

x sin nx dx = J -[~]
From (8.21) it then follows that

2 1 . 1
(-n + 1)a 1 = - - t.e. a1 = ~0.0564 (8.57) u (x)
2 2(n 2 - 1)
0.10 Subdomain col l ocat ion
For the subdomain collocation method, K and fare given by (8.33). In the present
""- Exac t
example, only one subdomain is involved and this is conveniently chosen as the entire
0.08 ;,----"
region of interest. This results in
0.06 /
/.
--· ".
// _ .... ' ---- . . . ."
//. -.........: .
'y
Ga l erk i n and l east squ ares

0.04 1. ~ ' ~
(8.58)
11/"~ '<0~
~
,/-/ Poi nt collocation '~
0.02 ~· '·
'~~

Use of these expressions in (8.21) results in ~------T-----~~----~------~-----x

0 0.25 0.5 0 .75 1.0
2 2 1 7t
(1-n ) - a 1 = -- i.e. a 1 = ~0.0885 (8.59)
1t 2 4( 1t 2 - 1) Figure 8.5 Comparison of different weighted residual methods
156 Introduction to the finite elem ent method

That is, (8.21) yields 9

2
1 2
- ( 1 - n )a 1
2
= - -1
1t
I.e. a1 = 2
n(n - 1)
~ 0.0718 (8.62) FE formulation of
It appears that the Galerkin and least-squares methods in this example provide
the same results. The results of the four weighted residual methods are compared
one-dimensional heat flow
with the exact solution in Figure 8.5 and the superiority of the Galerkin and
least-squares methods is obvious for the example considered.

\ 8.6 Concluding remarks

In the previous chapters we have discussed strong and weak fo rms of the heat flow
The point collocation method requires that the residual is equal to zero at the
problem, the elementwise approximation technique used in the FE formulation for
collocation points. This condition seems intuitively less exacting than that used in
the unknown function and various choices for the weight function. It was emphasized
the subdomain method where the average of the residual over each subdomain is
that the FE approach is based on the choice of weight function in accordance with
required to be zero. Thus one may expect the subdomain method to be more accurate
the Galerkin method. The basic steps in the FE method can therefore be summarized
than the point collocation method. The least-squares method and the Galerkin
as follows:
method turn out to be very efficient. The least-squares method always results in a
symmetric coefficient matrix ( cf. (8.41 )), which is clearly an advantage in numerical
calculations. In general, the coefficient matrix ( 8.45) in the Galerkin method is not
symmetric, so that the advantage of using this method in the FE formulation may Basic steps in the FE formulation
not seem obvious at this point. However, the weighted residual methods discussed 1. Establish the strong formulation of the problem.
in this chapter are based on the integral formulation (8.4) of the differential equation 2. Obtain the weak form of the problem.
(8.2). We have previously stressed that (8.4) is not a weak formulation. In a weak 3. Make an elementwise approximation over the entire body of the
formulation, an integration by parts is performed whereby the order of differentiation unknown function.
of the unknown function is decreased at the expense of the weight function being 4. Choose the weight function in accordance with the Galerkin method.
differentiated (see for instance (4.36)). As shown in the next chapter, this implies that
when the Galerkin method is used in combination with the weak formulation, a In order to illustrate these steps we shall consider the FE formulation of
symmetric coefficient matrix arises. This aspect as well as the applicability of this one-dimensional heat flow. Even though this constitutes a simple physical problem,
method to any differential equation suggest its fundamental importance in FE it will turn out that its FE formulation contains all the fundamental fea tures of any
formulations. FE approach. Therefore, the reader is urged to read the present chapter especially
carefully. If the procedures described here are fully comprehended, it will be an easy
task to follow the remaining exposition which discusses the FE formulation of more
complicated physical problems, but where no essentially new FE concepts will be
introduced.

9.1 Global FE formulation

Consider one-dimensional heat flow in the fin shown in Figure 9.1. The strong form
of a quite similar problem was given by ( 4.8 )-( 4.10) and with the position of the fin

157
158 Introduction to the finite element method FE formulation of 1-D heat flow 159

T The approximation for the temperature T is now introduced. Following the

summary of Chapter 7, the approximation over the entire region is in general written as

T=Na (9.4)

where N is the global shape function matrix and a contains the temperatures at the
A(x)
nodal points in the entire body (cf. (7.137) and (7.138 )). That is, we have

'-----~!::=::====:r--x
0 Tl
b
Tz
WIIJJILLD Q (x)
N = [N1 N2 .. . Nn]; a= (9.5)

T,
Figure 9.1 Heat conduction in one-dimensional fin
where n denotes the number of nodal points for the entire body and N ; = N ;(x ). As
as shown in the figure, the differential equation becomes a does not depend on x, (9.4) implies that

~
-d ( Ak -dT)
dx dx
+ Q= 0; a:::;; x :::;; b (9.1 )

where it is recalled that A(x) is the cross-sectional area of the fin, k(x) is the thermal
conductivity and Q(x) is the heat supply per unit time and per unit length of the fin.
t.e.
08 X
where
~
(9.6)

Q may be transferred to the fin across its outer surface or created internally, for B = [dNI dN2 .. . dNn]
instance by electric heating. In order to solve (9.1 ), boundary conditions must be (9.7)
dx dx dx
specified, for instance in the form given by ( 4.9) and ( 4.10), but to be able to deal
with completely arbitrary boundary conditions, we shall leave the boundary conditions
unspecified until later.
To obtain the weak form of (9.1) we multiply by the arbitrary weight function
v( x) and integrate over the region of interest, i.e.
AkB dx)a = - [vAq]~ + r
in accordance with (7.140) and (7.141). Inserting (9.6) in (9.3) we obtain

(f ~: vQ dx (9.8)

r{ ddx ( Ak ~:) + QJdx = 0

The first term is integrated by parts in accordance with ( 4.33) to provide
It is important that the approximation for d T I dx as given by (9.6) is not inserted
into the boundary terms on the right-hand side of (9.8), which include the flux
q = - k d T I dx. The reason is that the boundary conditions either specify the flux or
the temperature itself at the boundary and there is no reason to make an
approximation for matters that we know beforehand. We shall return to this topic
f bdx
a
dv Ak dT dx = [vAk dT]b + fb vQ dx
dx dx a a
(9.2) later on.
The last step is the choice of weight function v. In accordance with the Galerkin
and using Fourier's law q = -k dTi dx, we get method, we choose the weight functions to be equal to the trial functions (cf. (8.42)
and (8.43)). In the present case the trial functions are the shape functions; that is,
b-dv Ak -dT dx = - [vAq]~ + fb vQ dx ( 8.13) yields
fa dx dx a
(9.3)

I v=Nc I (9.9)
It appears that the integration by parts implies that the order of differentiation for
T has decreased at the expense of the weight functiort v being differentiated. Since v is an arbitrary function, the matrix c is arbitrary. Moreover, as v is one
160 Introduction to the h'nite element method FE formulation of 1-D heat flow 161

number, we have v = vT, i.e. (9.9) can be written as and defining the force vector f by
V = cTNT (9.10)
1 f = rb + f1 (9.1 5)
which implies that
(9.14) can be written as
d dNT
_...!:. = cTBT where BT =- - (9.11 )
dx dx I Ka =f I (9. 16)
Inserting (9.10) and (9.11) into (9.8) and using the fact that cT is independent of x
The temperatures at the nodal points given by a are obtained by solving this system

r
results in
of linear equations. When a is known, the temperature T at an arbitrary point in the
cT[ (f BT AkB dx )a+ [NT Aq]~- NTQ dx J= 0
fin is given by (9.4) and the temperature gradient at an arbitrary point is given by
(9.6). From this temperature gradient, the flux qat any location in the fin is obtained
from Fourier 's law (4.3). Therefore, when a has been determined from (9.16) all

(f BT AkB dx )a=- [NT Aq]~ + r

As this expression should hold for arbitrary cT-matrices, it is concluded that

NTQ dx (9.12)
quantities of interest can be derived. Let us now evaluate the stiffness matrix K and
the force vector f in more detail.

which is the required FE formulation. Here, we have derived (9.12) by means of(9.8)
and (9.9) and using the fact that vis arbitrary, and thus cis also arbitrary. Alternatively,
in (9.8) we may directly choose n arbitrary weight functions v and first choose v = N 1 ,
then v = N 2 and so on. Clearly, the result of this approach is again given by (9.12).
KT = ( r
From (9.13) and (2.30) it follows that

BTAkB dx)T = r BT AkB dx = K

i.e. the stiffness matrix K is symmetric. It is obvious that this symmetry hinges on
(9.1 7)

This discussion is similar to that relating to (8.16). Here, we prefer the approach the use of the Galerkin method. If instead of this method (9.9) we had chosen v = Vc,
given by (9.9) as it facilitates a clear matrix formulation, especially in the more where V -:f. N , then the corresponding stiffness matrix would have taken the fo rm
complicated FE formulations that will be encountered later. However, one should
bdVT dVT
be aware of the equivalence of the two approaches.
In order to write (9.12) in a more compact fashion, we define the following
matrices:
K =
J -
a dx
AkBdx where --:f. BT
dx
which implies that the symmetry of K would have been spoiled. This symmetry, the

K= r BTAkBdx
applicability of the Galerkin method to any differential equation and the realistic
predictions provided by this method are the primary reasons for its ado ption in the
FE formulation.
To evaluate the character of the stiffness matrix K further we consider its
= -[NT Aq]~

r
fb (9.13)
components. From (9.13) and (9.7) it follows that

fl = NTQdx 1 kdN 1 d 1 kdN 2 d

rdN--A -- X rdN--A -- X rdN1 kdN"d
- -A - X
a dx dx a dx dx a dx dx
Referring to (9.7) it is obvious that K is a square matrix with dimension n x n; it is rdNz k dN"d
r dN 2 Ak dN 1 dx r dN 2 Ak dN 2 dx --A - X
called the stiffness matrix. Likewise, both fb and f1 have the dimension n x 1, and K= a dx dx a dx dx a dx dx
they are called the boundary vector and load vector, respectively, since fb refers to
conditions at the boundary, whereas f1 considers the effect of the 'loading' Q. With
(9.13), (9.12) takes the form
r dN" Ak dN ! dx rdN"
- AkdN
- -2 d X r -dNn
AkdN"d
-- X
a dx dx a dx dx a dx dx
(9.14)
(9.18)
162 Introduction to the finite element method
FE formulation of 1-D heat flow 163
ol
N· 2 3 5 6
I~

.l\
Figure 9.3 Fin modelled by five linear elements

stiffness matrix becomes banded with non-zero components clustered about the
diagonal of K. This has important numerical advantages as discussed in Chapter 11.
We emphasize that the banded structure of the stiffness matrix K is a result of the
b) fundamental property that a global shape function differs from zero only in those
elements which contain the nodal point in question.
As an illustration of these features, consider the fin shown in Figure 9.3. The fin
is modelled by five elements and for convenience we make use of the simple linear
element of Chapter 7. With the adopted nodal numbering it follows that the stiffness
Q Q matrix K takes the following fo rm:

1 2 3 4 5 6
elem. elem. elem .
• I • • I

1 1 1 1
X X 0 0 0 0
Figure 9.2 Variation of global shape functions N; and Ni for linear elements 2 X X X 0 0 0
3 0 X X X 0 0
K =4
The symmetry of K is also obvious from this expression. Moreover, consider a 0 0 X X X 0
component Kii of K, which can be written as 5 0 0 0 X X X

f
bdN- dN . 6 0 0 0 0 X
K .. = - ' Ak - 1 dx (9.19) X
I] a dx dx
In order to evaluate the force vector f given by (9.15), let us first consider the
The approximation of the temperature according to (9.4) is an element":ise
components of the boundary vector fb and the load vector f1• F rom (9.13) we-get
approximation, and in order to evaluate (9.19) we recall that one global shape functiOn
is related to each nodal point. The variation of the global shape functions N; and
N . is shown in Figure 9.2, where, for simplicity, it has been assumed that the simple
li~ear finite element has been used, i.e. N; and Ni vary linearly ( cf. Figure 7.8).
However, the important point, which is common to all types of finite elements, is
that a global shape function N; differs from zero only in those elements which contain
the nodal point i. As shown in Figure 9.2, two principally different possibilities exist (9.20)

r
for the variation of the global shape functions N; and Ni. In Figure 9.2(a), no elements
contain both nodal points i and j , whereas in Figure 9.2(b) one element contains
both nodal points i andj. In Figure 9.2(a), N; is zero when N i is different from zero [N.AqJ! N. Qdx
and vice versa, i.e. when evaluating (9.19) the component Kii is zero. Only when the
two global shape functions N; and Ni differ from zero at some common regions will
A typical term fbi of the boundary vector fb is given by
K 11.. be different from zero. Such an example is shown in Figure 9.2(b) wpere N; and
N only differ from zero simultaneously in the element given by nodal points i and fbi= - [N;Aq]~ =- (N;Aq)x=b + (N; Aq )x=a (9.2 1)
j. In conclusion, the component Kii of the stiffness matrix K is zero unless both nodal
The positions x = a and x = b, i.e. the ends of the fin, clearly correspond to the
points i and j are present in at least one element. This conclusio~ i.s si~ilar to the
position of two nodal points. For the sake of illustration assume that the nodal points
observation made in Chapter 3 (cf. (3.70) and (3.71)). Moreover, tt tmphes that the
1 and n are located at x = a and x = b, respectively. With the properties of a global
164 Introduction to the finite element m ethod FE formulation of 1-D heat flow 165
shape function in mind it follows that 2 3 ~

N 1 (x =a)= 1 if i = 1
~----~====i=====~====~----- X
(9.22) X : 2 = I. =6 :8
N -(x =a)= {
I 0 if i -# l
Figure 9.5 Three linear elements of equal Length
and where 9a and hb are known quantities. These boundary conditions correspond to a
N.(x =b)= 1 if i = n prescribed temperature at the left end of the fin and a prescribed flux q = - k d T / dx
N -(x =b) = { (9.23)
1
0 if i-# n at the right end. Moreover, let

With (9.20)-(9.23), the boundary vector becomes ga=0 °C ; hb = l5J/ m 2 s (9.28)

Assume that we divide the fin into three elements of equal length and also assume
(Aq)x=a
for convenience that we adopt the simple linear element of Chapter 7. The nodal
0 points are indicated in Figure 9.5. The global shape functions together with their
(9.24~ derivatives are shown in Figure 9.6.
Let us first determine the stiffness matrix K. From (9.19) it follows that
0
- (Aq)x=b Ku = fb d N, Ak dNI dx =
a dx dx
J4 dx Ak dx dx
2
d N, d N,

I
where q is the flux given by Fourier's law (4.3).
4
= ( -0.5)50( - 0.5) dx = 25
9.1. 1 Example 1
where N 1 , i.e. dN 1 / dx, o nly differs from zero in the interval 2:5: x :5: 4. Likewise
Having discussed the FE formulation of one-dimensional heat flow in general terms,
it is timely to investigate the solution of a specific example. For simplicity, we assume
that the fin shown in Figure 9.4 has a constant cross-sectional area A, a constant
- f b-
K 12- dN-I A k -
a
dN-2 dX--
dx dx
J 2
4
dN
-
dx
1 k dN 2 d
-A -- X
dx
thermal conductivity k and a constant heat suppl.y Q.
From F igure 9.4 and (9.1) the differential equation is given by = I 4
( - 0.5)50(0.5) dx = - 25

~(Ak dT) + Q = 0; a :5: x :5: b (9.25) Here, we used the fact that N 1 and N 2 onl y differ from zero simultaneously in the
dx dx range 2 :5: x :5: 4. Since N 1 and N 3 have no common region where they differ from
zero, we obtain
where
bdN 1
f
A= l0m 2 ; k=5 J ;oC m s; Q = lOOJ / sm; a =2m ; b=8m (9.26) dN
K1 3 = --Ak - -3 dx = 0
a dx dx
The boundary conditions are assumed to be of the form
and likewise
(9.27)
dN
f
b dN 4
K 14 = - -1 Ak--dx =0
a dx dx
~A=1 0.k=5
AsK is symmetric (cf. (9. 17)), we next consider component K 22 , i.e.

a= 2 b= 8
• X

K 22 = bdN2
- A k -dN2 dx =
f a dx dx
I 6

2
dN2
- A k dN2
dx
-dx =
dx
J 4

2
(0.5)50(0.5) dx

f f f f f f f f f f f f fa=1o o
Figure 9.4 Problem definition
+ t 6
( - 0.5)50( - 0.5) dx = 50
166 Introduction to the finite element method
FE fo r mulation of 1-D heat flow 167

f6
X=2 =I. =& =8 dN 1 /dx x=2 :I. =6 =8
Nt
K 33 = Jb dN 3
Ak dN3 dx = (0.5)10 x 5(0.5) dx
dx dx

'f
a 4

~; 3
§
{.
O--x
0.5 2 3 {.
X
+ L 8
( -0.5)10 x 5( -0.5) dx =50
- 0. 5 t-
8
K 34 = fb dN 3
Ak dN 4 dx = f ( -0.5)10 x 5(0.5) dx = - 25
N2
dN2 idx a dx dx 6

K 44 = Jb dN 4
Ak dN 4 dx = f 8
(0.5) 10 x 5(0.5) dx = 25
'f !4':>~ I.
~X
0.5 2 3 f.
X
a dx dx 6

Collecting these results together, we obtain the stiffness matrix K:

-0.5
\ 25 -25 0 0
-25 50 -25 0
N3 dN/dX K= (9.29)
0 -25 50 -25
0 0 -25

'i
25
Q
2~1.O--x
§ 0
0.5 z---3 {.
X The banded structure of the stiffness ma trix is obvious. Next, consider the boundary
-0.5 vector fb. From (9.24) and (9.26)- (9.28) we obtain

(Aq)x ca lOq(x = 2)
N, dN4 /dx
0 0
fb = (9.30)

't
0 0
1
Q
2
§
3/.
§ o----x
0.5 2
.2.
3
X
- (Aq)x ; b -150

r
-0.5 The load vector f1 is given by (9.20), and from (9.26) and F igure 9.6 we find
4
Figure 9.6 Global shape functions and their derivatives N 1 Qdx N 1 dx

r
{ 100

Moreover, proceeding in the same manner we get

r, =
N 2Qdx {
4

N 2 dx + L6

N 2 dx 200

L
=Q (9.31)
K = fb dN2 Ak dN3 dx = f6 dN2 Ak dNJ dx Lb N 3Q dx
L 6

+
8
200

r
23 a dx dx 4 dx dx N 3 dx N 3 dx

= t6
( -0.5)50(0.5) dx = -25 N4Qdx L 8

N 4 dx 100

and where again we utilized the fact that a global shape function Ni differs from zero
only in those elements wruch contain the nodal point i.
K 24 =
fa
bdN 2
-Ak -
dx
dN 4
dx
dx=
O
From (9.29)- (9.31), the final equation system (9.16) can be established and since
the prescribed temperature T1 is T1 = T(x =a)= ga = 0 {cf. (9.27) and (9.28)), this
168 Introduc tion to the finite element method FE formulation of 1-D heat flow 169
implies that T

25 -25 0 0 0 10qx=2 100

20
-25 50 -25 0 T2 0 200
0 -25 50 - 25 0
+ (9.32)
T3 200
0 0 -25 25 T4 - 150 100
r;
It can be noted that the largest computational effort is involved in determining 10 'I
the stiffness matrix K and that other flux boundary conditions and loads Q only '/--FE solu tio n
influence the right-hand side of (9.32). Considering the first row and the last three '/
rows of (9.32) separately, we obtain 'I
'I
- 25T2 = lOq(x = 2) + 100 (9.\) ~~===r==~==~-- x
2 I, 6 B
and
dT / dx
-25
[ 50
-2~ 50 OJn [ 200J
- 25 T3 = 200 (9.34)
10
-25 25 T4 - 50
By using the Gauss elimination technique presented in Chapter 2, it can easily be
shown that the solution of (9.34) becomes
5

(9.35)

With the known T2 -value, the flux qat x = a= 2 is determined from (9.33) to provide
-1
qx=2 = - ( k -dT) = - 451/ m 2 s (9.36)
dx x = 2
i.e. heat must be extracted at the left end of the fin to keep the temperature at the
required value. Figure 9.7 Comparison of exact and FE solution
To evaluate this solution, we make comparisons with the exact solution. It can
easily be checked that by Hughes ( 1987, p. 27)). Besides, a fair agreement exists between the exact and the
FE solution and we may draw the general conclusion tha t, even though the
T = - x 2 + 13x- 22
. (9.37)
is the exact solution which satisfies the differential equation (9.25) and the boundary
temperatures of the FE solution are close to the exact values, the temperature gradient
of the FE solution is in much less agreement with the exact temperature gradient.
conditions (9.27) with the specific data given by (9.26) and (9.28). In recognition of this, it may be of some surprise that the flux q at the left end of
The FE solution and the exact solution are compared in Figure 9.7, where due the fin as determined by (9.36) is in perfect agreement with the exact solution, as is
to the simple element adopted, the temperature determined by the FE method varies easily checked from (9.37). The reason is because the temperature gradient dTj dx
linearly between its values at the nodal points. It is of interest that the FE solution present in (9.36) is not determined from dT j dx = Ba, which would lead to an inferior
is exact at the nodal points. answer, but rather from the boundary vector fb where the exact q-value enters.
While this is certainly not a general conclusion, we note that this fortunate In a later section we shall prove that the balance principle for the entire body is
situation occurs for some one-dimensional problems (a detailed discussion is presented fulfilled exactly by the FE solution, and as the loading due to Q is given, as is also
170 Introduction to the finite element method
FE form ulation of 1-D heat flow 171

the flux at the right end of the fin, the FE solution is forced to provide the exact flux To evaluate the properties ofthe stiffness matrix K further, we make a partitioning
of K and a (cf. Chapter 2). As an example we split K and a into the following
at the left end of the fin.
submatrices:

(9.44)
9.1 .2 Further properties of the stiffness matrix
In this partitioning ofK, use is made of the fact that K is symmetric and it also follows
Apart from the problem d iscussed in Example 1, the boundary conditions rela~ing that the submatrix ~ is symmetric, i.e.
to the differential equation (9.1} have not been discussed , i.e. the system of equ~t~ons
(9.16) holds fo r arbitrary boundary conditions. Clearly, the boundary conditions (9.45)
must be considered in order to solve a specific physical problem and in o rder to do
so in a systematic manner we first need to establish some important properties of Using (9.44) the quadratic form (9.41) can be written as
the stiffness matrix K. \
We shall first show that f=[gT ~TJ[~i ~2 J[:J;;;::o (9.46)

I det K =0 I (9.38)
where (9.43) has been used. We emphasize that the submatrices given by (9.44) must
have proper dimensions so that the matrix multiplications indicated in (9.46) are
To prove this, consider the following homogeneous equation system: permissible. Evaluating (9.46) we obtain
Ka = 0 (9.39)
(9.47 )
Referring to (2.55), we recall that a non-trivial a-solution exists if and only if (9.38)
Taking account of (9.43), we observe that I= 0 if and only if all components of the
is true. As a is independent of x, use of (9.13) in (9.39) implies that a-matrix are identical; otherwise we have I > 0. Therefore, choosing aT = [gT ~ T] =

f BT AkBa d x = 0 (9.40)
[OT ~ T] it follows from (9.47) that

All the finite elements considered are able to reflect a zero temperature gradient
I ~ TK~ > 0 I (9.48)

exactly (cf. (7.2)). Therefore, if all the temperatures at the nodal points are equal ~h~n for~ =ft 0. This proves that K is positive definite and from (2.67) it then follows that
dTj dx = Ba = Oeven though a =ft 0, i.e. (9.40) and thus also (9.39) possess a non-tnv1al
a-solution. In tum, this proves that (9.38) is correct.
Next, consider the quantity I defined by the following quadratic form :
1 det ~ =ft o 1 (9.49)

The submatrix K considered above can be viewed as o btained from the stiffness
(9.41) matrix K by deleting some rows and corresponding columns in the manner indicated
by the partitioning shown in (9.44 ). It then turns out that K possesses the three
With (9.13) and (9.6) we get fundamental properties given by (9.45), (9.48) and (9.49). However, by using exactly
the same type of arguments as presented above, it follows t hat any submatrix :{S:
I = fb a TBT AkBa d x = fb dT Ak dT dx ;;;:: 0 .
(9.42) obtained from the stiffness matrix K by deleting one or more rows and the
a a dx dx
corresponding columns possesses these fundamental properties. This conclusion bas
where Ak > 0 and (dTj d x) 2 ;;;:: 0. As I ;;;:: 0 for arbitrary a-matrices different from important consequences when considering the boundary conditions, as shown next.
zero, we conclude from (2.68) that the stiffness matrix K is a positive semi-definite
matrix. It also follows from (9.42) that
9.1.3 Systematic consideration of boundary conditions
I > 0 ifdTj dx =ft 0 (9.43) When the system of equations (9.16) was established, no use was made of the boundary
I = 0 if d T / dx = 0 conditions. The differential equation (9.1) contains only the temperature gradient
172 Introduction to the finite element method FE formulation of 1-D heat flow 173
dT/ dx. Let us assume that the boundary conditions are prescribed in terms of given indicated in (9.54 ). Let us define the following submatrices of the stiffness matrix K:
fluxes at both ends of the fin, i.e.
At= [KttJ; A2 = [Kll K 13 K 1,.]
(9.50) Kzt K22 Kz3 Kz,.
where ha and hb are known quantities and where it is recalled that q = -k dTj dx. K31 K32 K33 K 3n (9.55)
AI= K=
In this case, both the differential equation and the boundary conditions involve only
the temperature gradient d T I dx. This means that if T(x) is a solution to the problem,
K. t K.z K,.J K.,.
then T(x) + C, where C is an arbitrary constant, is also a solution to the problem.
In this situation, we would expect that a unique solution to (9.16) does not exist. where the symmetry of K was used in the definitions of A 2 and AI . Likewise, we
Referring to (2.56), we would therefore expect the determinant of the stiffness matrix define the following submatrices of a and f:
K to be zero and this is precisely the situation indicated by (9.38).
Tz /2
Consequently, in order to obtain a unique solution from (9.16) we must s~cify
at least one nodal temperature, and valid boundary conditions could therefore ~ of T3 !3
the form g = [ga]; ~= r = UtJ; f= (9.56)

q(x =a)= ha; T(x=b)=gb (9.51) T,. /,

T(x =a)= ga; q(x =b)= hb (9.52)
This means that the system of equations (9.54) can be written as
T(x =a)= ga; T(x=b)=gb (9.53)
(9.57)
where h0 , hb and ga, gb are given values of the fluxes and temperatures, respectively.
We shall now prove that these boundary conditions lead to a modification of the
It appears that the only unknowns in this equation system are ~ and r. Performing
system of equations (9.16), which ensures a unique solution. The proof follows the the matrix multiplications indicated in (9.57) we obtain
same lines irrespective of whether (9.51), (9.52) or (9.53) is given and, as an example,
we assume boundary conditions in the form given by (9.52). For convenience, we A 1g+ A2 ~ = r
also assume that nodal points 1 and n are located at x = a and x = b, respectively;
Aig + K~ = f
that is, at x =a we have T1 = ga. With (9.15) and (9.24), the system of equations
(9.16) can be written as which can be restated as

I
Kll I K12 K!3 K~n ga fu ft
K~ = f- Aig I (9.58)
----,----- ------- --(Aq)x= a
-- - --
K21 1 K22 K23 ··· K2. T2 0 j,2 fz
K31 : K 32 K33 K3n T3 0 + j,3 /3
r = Atg + Az~ I (9.59)
I
I The right-hand side of (9.58) is known and asK is obtained from the stiffness matrix
K,.t : K,.z K,.J K,.. T,. -A(x = b)hb f,. /, K by deleting a row and the corresponding column, Kpossesses the properties given
by (9.45), (9.48) and (9.49). In particular, det K =F 0, i.e. a unique solution l! is obtained
(9.54) from (9.58) and this solution does not depend on the unknown r-matrix. By inserting
the ;!-solution into (9.59), we can determine the unknown r-matrix. Therefore all
All the components of the stiffness matrix are known and the same holds for the quantities of interest have been calculated in a unique manner. We also take' the
components of the load vector f1• However, the total amount of beat flow at x =a, opportunit~ to draw attention to the similar observations made in Chapter 3 regarding
i.e. (Aq)x=a> is unknown, which means that the force component / 1 is unknown, the properties of K and K as well as the introduction of boundary conditions ( cf. the
whereas the components / 2 .. .f,. are known. Moreover, component ga of a is known, discussion related to (3.25)-(3.32)).
whereas the components T2 ... T,. are unknown. A partitioning of the matrices is The systematic consideration of the boundary conditions illustrated above proves
174 Introduction to the finite element method FE formulation of 1-D heat flow 175
that, when at least one nodal temperature is prescribed, a unique solution of the
system of equations (9.16) follows. Moreover, this systematic approach lends itself 200]
directly to a computational implementation of boundary conditions in an FE program.
r = [ 10q(x = 2) + 100]; f = 200 (9.63)
With the discussion above we are even able to find all solutions to (9.16) when [ -50
the boundary conditions are given in terms of prescribed fluxes alone ( cf. (9.50)). For From (9.58) we then obtain
this purpose we prescribe an arbitrary temperature at one end of the fin, for instance

[ -~~ -;~ -2~][~]=[ ~:]-[- ~][0] =[ ~:]

at x =a; that is, we set T(x = a) = ga, where ga is arbitrary. Assuming that nodal 2
point 1 corresponds to x =a, we have T(x =a)= ga = T1 , i.e. we again obtain the
system of equations (9.57). Now, the only unknown matrix is~. but this matrix can
be found by solving (9.58), i.e. all nodal temperatures are now available. If we denote 0 -25 25 T4 -50 0 -50
this solution by T, it is obvious that T + C is also a solution, where C denotes an This system of equations is identical to (9.34 ), i.e. the solution becomes
arbitrary constant.

\ (9.64 )

9.1.4 Example 2
From (9.59) and (9.61)-(9.64) we get
In order to illustrate the systematic consideration of boundary conditions as described

'~ ~ 2) + 100] ~ (25](0] + [ -25 o{~~] ~

above, we again consider the problem of Example 1. According to (9.14) we have
seen in general that [!Oq(x 0 -350 (9.65)
Ka = fb + f1
For the problem of Example 1 this system of equations takes the form (9.32), i.e. i.e.
I
25 1-25 0 0 0 10qx=2 100 q(x = 2) =- 45 (9.66)
- ---r- - - - - - - ---- --- ------
-25 I 50 -25 0 Tz 0 200 The solutions (9.64) and (9.66) are evidently in accordance with (9.35) and (9.36) of
I
+ (9.60) Example 1. Even though the analysis above may seem a little involved, the important
0 1I - 25 50 - 25 T3 0 200
point is that we have illustrated how a systematic consideration of the boundary
0 : 0 -25 25 T4 -150 100 conditions can be carried out. Moreover, this systematic approach applies to general
where the specific boundary conditions were introduced. The unknowns of this situations, i.e. it lends itself directly to an efficient computer implementation of the
equation are T2 , T3 , T4 and 10q(x = 2). To solve it in a systematic manner, we make FE method.
the partitioning as shown above, which is analogous to that indicated by (9.54 )- (9.57).
That is,
9.1.5 Evaluation of the force vector- fulfilment of the global
A1 = [25]; A2 = [ -25 0 OJ balance principle
(9.61)

-2~]
-25
[ -25 ] [ 50 It is an important characteristic of the FE method that even though it is an
A1 = ~ ; ~ = -2~ 50 approximate method, the balance principle for the entire body is fulfilled exactly. In
-25 25 Figure 9.8 all heat quantities which enter and leave the one-dimensional fin are shown.
and As the flux q given by Fourier's law (4.3) is positive in the x -direction, (Aq)x=a is

g~(O]; ·~m
the heat per unit time which enters the fin at end point x =a and (Aq)x=b is the heat
per unit time which leaves the fin at end point x = b. We recall that Q is the heat
(9.62)
input per unit time and per unit length of the fin and is measured as positive if heat
is supplied to the fin.
176 Introduction to the fini te element method
FE formulatiOn of 1-D h eat flo w 177

IAq)~.a-_1_ _ _ _ _ _ _ _ _]-IAq)~ = b This means that the exact fulfilment of the balance principle for the entire body is
expressed by the fact that, in accordance with (9.68), the sum of all the components
of the force vector f is equal to zero. As an illustration consider the solution obtained

WlL1IDliJ Olx)
in Examples 1 and 2. F rom {9.60) and (9.66) it follows directly that the force vector
fulfils the balance principle of (9.68). Recalling that our FE equations were derived
from the balance principle itself, the property (9.68 ) is indeed not surprising.
~----~----------+-------x
a b
9.1.6 Evaluation of the load vector- point sources
Figure 9.8 Heat supply and removal over the entire body
Let us continue the ev·aluation of the force vector f. For this purpose we will provide
As we are considering time-independent problems, the balance principle for the a physical interpretation of the boundary vector fb and the load vector f1• We o bserve

(Aq)x=a + r
entire body states that the heat entering the body equals the heat leaving the body, i.e.

Q dx = (Aq )x=b
that they have the same dimension, and to establish this dimension it is convenient
to consider f 1 as given by (9.13). The global shape function matrix N is dimensionless,
and as Q is the heat supply per unit time and per unit length of the fin, then f1 and
thus also fb have the dimension [J/ s ] , i.e. heat per unit time.

r
which can be written as In o rder to obtain a physical interpretation of the load vector f1 we evaluate a
component J.1 of f 1. From (9.20) such a component is given by
- [Aq]~ + Qdx= O

This equation is the balance principle for the entire body. Let us consider now.the
force vector f which, according to (9.15), is gi ven by
{9.67)
/. 1 = r
N 1Q dx {9.69)

It follows that if the load Q(x) is prescribed over a region where N 1 is different fro m
ft=/b;+/li; i= l , ... , n zero then fu # 0, otherwise f.1 = 0.
Next let us consider the influence of a load Q in terms of a so-called point source.
This implies that This means that the heat supply is concentrated to a point and at this point a heat
n n n supply Q. per unit time is prescribed. Q. is called the strength of the point source
L: !t = L: fbi + iL:=- /.;
i ::: 1 i= 1 1
and it has the dimension [J / s]. Using Dirac's delta function (cf. (8.23)- (8.25)), we
may express the load Q as
From (9.20) we get
ifx = c
Q = Q5 b (X - C) = { : (9.70)
n n n f b
~~~ /; = - i~t [N,Aq]~ + ~~~ a N,Q dx
otherwise
where c is the position of the point source. By definition, we have that

=- [(.t 1-l
N,)Aq]b + fb
a a
(.± )Q
1= 1
N1 dx
J~co Q dx = J~oo Q,<5(x- c) dx = Q. (9.71)
Use of (7. 139) yields
Due to (9.70), (9.7 1) may also be written as
fb
Lh= - [Aq]~ +
n

i= ! a
Qdx
f_' Q.<5(x - c) dx = Q. {9.72)
and from (9.67) we conclude that
where c + and c- denote x -values slightly larger than and smaller than c, respectively.

I J,!.~D I (9.68)"
/.1 = r
For such a point sour ce, the component /. 1 given b y (9.69) becomes

N 1Q5 <5(x- c) dx = f_· N,Q.D(x- c) dx = N 1(c) f_·

Q.D(x- c) dx
178 Introduction to the finite element method FE formulatiOn of 1-D heat flow 179
i.e.
fin. From (9.15) we have
n n n
(9.73) I ;; = I
i=l l= i
fbi + I
i =l
J;, (9.76)

If the position c of the point source coincides with the position of nodal point i, then •
Using (9.75) and Figures 9.9 and 9.10, it is obvious that the balance principle states that
N;(c) = N;(x;) = 1, i.e. we have
n

J;; = Q. if the point source is located at nodal point i (9.74)

I;; =o
i= I

in accordance with (9.68).

Ref~rring .to (9.69) for an arbitrary load Q, it follows that it is always possible to
devise pomt sources located at the nodal points and with such strengths that J;
obtained from (9.74) equals J;; obtained from (9.69), i.e. u

J;; = f N;Q dx = Q,; (9.75)

9.2 Expanded FE formulation of one element-
where Q.; is t?~ strength of the nodal point source located at nodal point i. We may assembling process
th~refore envision the components of the load vector f 1 as given by the strengths of
~om~ sources located at the nodal points. This equivalence is shown schematically Previously, we established the FE formulation for the entire body, but for computer
m Figure 9.9. As the load Q is positive when heat is supplied to the fin, the same implementations it will turn out to be important also to know the FE formulation
holds for the strengths Q,; (cf. Figure 9.9). Occasionally, the components of the load of one element. The number of such an element is deno ted by a and the ends of the
vector f1 from a distributed load Q are obtained by intuitive physical considera1ions. element are located at x = X; and x =xi, as shown in Figure 9.11. For the linear
However, when determining f1 using (9.69) one speaks of a consistent load vector element the only nodal points are those at x = X; and x = xj, but for higher-order
since this determination is consistent with the Galerkin method. ' elements we also have interior nodal points. The space which element a occupies is
Let us now return to the boundary vector fb given by (9.24 ). In accordance with denoted by L~~. .
Fouri~r's law (4.3), the flux q is measured as positive in the direction of the x-axis. The approximation of the temperature given by (9.4) holds for the entire body,
That 1~, the.components of fb given by (9.24) can be illustrated as in Figure 9.10. and therefore it also holds in the region La.. By treating the region La. as a separate
With F1gures 9.9 and 9.10 we have now illustrated all the heat supplied to the body, we can make an FE formulation for this region and will clearly end up with
a formulation similar to (9.12). Noting that the region La. of element a is X;~ x ~xi
(cf. Figure 9.11 ), it follows that
1
9
2
0 § 0
n
9 ¢:::>
1
9
2
0 § 0
n
9 (I 1
BTAkB dx )a = - [NT Aq]~1 + J: 1
NTQ d x (9.77)

WlJID CIJ]J Q 0 s2
When performing integrations over a region La. which extends from X; to xi, it is
Qsl Qsn

Figure 9.9 Equivalence between distributed load Q and point sources Q,;
region La.

2 n X·
J
(Aq)x:o - 9'=~o==:JoC
'- - 0 -==::9--x

el em ent no.a.
~---r------------------~------x
a b
Figure 9.10 Illustration of boundary vector fb Figure 9.11 One element with ends at X; and xi and with possible interior nodal p1ints
180 Introduction lo the finite elemen t method FE formulation of 1-D heat flow 181
'
convenient to write this as region La, we therefore have

t 1

H(x) dx = L. H(x) dx (9.78) f H(x) dx = atl (L. H(x) dx)

where, again, H(x) is an arbitrary function. With (9.13), (9.80) and (9.84), we obtain
(9.84)

where H(x) denotes any function of x. Likewise, we adopt the notation

[NT Aq]~1 =[NT AqJr.. (9.79)
If we define the following matrices: (9.85)
n,

Kee = L. BT AkB dx
rl = L: ft'
a =l

fb• =-[NT Aqk (9.80) where K!• denotes the expanded element stiffness matrix for element a and ft" is the

ff• = f/...,
NTQ dx
expanded load vector for element a.
Therefore, the (global) stiffness matrix K is the sum of all the expanded element
stiffnesses K!". Likewise, the (global) load vector f 1 is the sum of all the expanded
element load vectors fj•. We emphasize that K and K!• have the same dimension
then (9.77) can be written as n x n, where n is the t~tal number of nodal points in the body. Likewise f1 and fl."
have the same dimension n x 1.
(9.81) With reference to Figure 9.12, let us consider two neighbouring elements with
numbers a and p. It appears that
"
For reasons that will be explained later, superscript ee denotes 'expanded element'.
(9.86)
That is, Kec is the expanded element stiffness matrix for element a, fb• is the expanded
element boundary vector for element a and ff• is the expanded element load vector for where q = - k dTf dx. We have seen that, in general, the temperature gradient dT/ dx
element a. Moreover, if we define the expanded element force vector r•• for element varies in a discontinuous manner over element boundaries; see for instance
a according to Figure 9.7. Considering point x j of Figure 9.12, it may therefore be somewhat surprising
that the boundary term from region L,. cancels the corresponding boundary term
(9.82) from region Lp . However, the important fact is that the boundary fluxes q entering
the boundary terms in (9.86) are the exact ones, since no approximation of the
(9.81) may be written as temperature gradient dT/ dx was used for these boundary fluxes (cf. the discussion
of (9.8)). Let us now determine the sum of the boundary terms -[NT Aq J~: for aU
I K••a = ree (9.83) the elements of the body.
From (9.86) and with the notation of (9.79), we obtain
This system of equations constitutes the FE formulation of one element. We remark
that as a in (9.83) is the same as that present in the system of equations (9.16), which -[NT Aq]~ = - "··
L [NT Aqk (9.87)
a= l
applies to the entire body, the dimension of a is n x 1, where n is the total number
of nodal points in the body. This means that the matrices K•• and f•• have the same
dimensions as those of K and fin (9.16), respectively. As (9.83) has been formulated ele men t a. el ement PI
in terms of all the degrees of freedom for the entire body and not only those related
I· I .
to the specific element, (9.83) is referred to as the ex panded FE formulation of one x ·I X·
J
xk

element. Examples of such a formulation are given by (3.15) and (3.17). =::2::=: ::=c: ~

Now Jet n. 1 denote the total number of elements for the entire body which occupies La L~
the region a ~ x ~ b; cf. Figure 9.1. An integration over this region can always be I·
expressed as a sum of integrations over each element. With element a occupying the Figure 9.12 Two neighbouring elements with regions La and L p
 FE form ulation of 1-D heat flow 183
182 Introduction to the fini te element m e thod

Using (9.80) and (9.13), (9.87) can be written as 9.2. 1 Example 3

In order to illustrate how the FE formulation of the entire body can be established
(9.88) as a sum of the contributions of all elements in accordance with (9.85), we consider
again the problem described in Example 1.
With this procedure we have to identify the number of the elements; thus
where ~ denotes the element boundary vector for element a.. Figure 9.5 is redrawn in Figure 9.13. As we have three elements, i.e. n.1 = 3, the
From (9.82) it follows that element numbers become 1, 2 and 3, but apart from that the specific numbering of
ncJ nd no::~ the elements if of course immaterial.
2:
a= l
r~· = I
a= l
fb~ + I ff."
a=l
We recall that the boundary vector fb is established in the same manner as before;
that is, from (9.30) we get
i.e. (9.85), (9.88) and (9.15) imply that 10q(x = 2)
0
n" (9.91)
r= 2: r:· (9.89) 0
a=l
- 150
We are now in a position to establish a fundamental property for the FE Next let us establish the expanded element stiffness matrix and expanded element
formulation. Based on the expanded FE formulation of one element, expression (9.83 ), load vector for each element. From (9.80), we have
we sum over all elements of the body to obtain

n cl nd
K•• =f
L.
BT AkB dx = f L.
dNT Ak dN dx
dx dx
I K:•a = L f!•
a= l a=l i.e. the component (K'•)ii of K•• is
dN- dN .
and with (9.85) and (9.89) this expression becomes

Ka = f (9.90)
(K••);· =
' f L.
-
dx
' A k -' dx
dx
Likewise, we have from (9.80) that a component (W )t of fj'• is
(9.92)

which is precisely the FE formulation of the entire body (cf. (9.16)). This shows that
the FE equations for the entire body can be established as the sum of the contributions
of all elements a nd the process of making this summation is called assembling. This
(fj'
0
)1 = f L.
N 1Q dx
(9.93)

essential result is analogous to the assembling procedure discussed in C hapter 3 Consider first element 1, where 2 :::;; x :::;; 4 corresponds to the region L 1 , i.e.
(see for instance the discussion of (3.23) and (3.26)). 4

f
dN- dN.
We emphasize that we need not establish the FE equations for the entire body (K~•)ii = -d'Ak -d ' dx
2 X X
as a sum of the contributions of all elements. We can just as well use the formulation
(9.16) directly, but it will turn out later that a slight reformulation of the process
given by (9.85) and (9.89) provides essential advantages when constructing a computer
program for FE analysis. We also note that obtaining the boundary vector fb as the
(fin = t 4
N;Q d x

sum of the expanded element boundary vectors fb in accordance with (9.88) does
not provide any advantages. In fact, from (9.24) it follows that the boundary vector e l.1 e l.2 el. 3
fb only possesses two non-zero components and it is therefore much more 2 3 4
advan tageous to establish the boundary vector fb directly from (9.24). Moreover, we
-C===±====i:;===:::9-=----
=6 =8
X
x=2
emphasize that the boundary conditions of an element are in fact unknown, except
when the element boundary coincides with the boundary of the body. Figure 9.13 Numbering of elements
184 Introduction to the finite element method
FE formulation of 1-D heat flow 185
A~cording to (9.26) we have. Ak = 50 and Q = 100. Moreover, referring to
Figure 9.6, the only shape f~n~t10ns which differ from zero in the region 2 .s; x < 4 The expanded element stiffness matrix K•• and the expanded element load vector
are N 1 and N 2 • It follows trlVlally that - ff• are of dimensions n x n and n x 1, respectively, where n is the total number of
nodal points. The nodal temperatures, which influence the temperature approximation
25 -25 0 0 100 within an element, are those related to the nodal points belonging to the element.
- 25 25 0 0 100 Therefore, if nodal point j does not belong to the element, the nodal temperature Ti
K••-
1- has no influence on the results and, consequently, the component (K••)ii of the
Cff = (9.94)
0 0 0 0 0 expanded element stiffness matrix is zero. As K •• is symmetric, we conclude that a
0 0 0 0 0 component ( K••)ii is zero unless both nodal point i and nodal point j belong to the
element a considered. Therefore, if the element contains n. nodal points then only
In the s~me manner for element 2, where 4 .s; x .s; 6 corresponds to the region L n. x n. components of the expanded element stiffness matrix K•• will be non-zero.
we obtam 2•
As K•• has the dimension n x n, this implies that many components of K•• are zero
0 0 0 0 0 (see for instance (9.94)-(9.96)). This will be especially pronounced if the body is
0 divided into many elements.
25 - 25 0 100
x··-
2 - ff;= (9.95)
As an example, assume that simple linear elements are used, i.e. n. = 2; then the
0 -25 25 0 100 number of non-zero components in each expanded element stiffness matrix is
0 0 0 0 2 x 2 = 4. Assume that the total number of nodal points in the body is n = 30. As
0
the dimension of K•• is n x n = 30 x 30 = 900, the number of zero components in
For element 3 with the region L 3 corresponding to 6 .s; x .s; 8, we find that K•• is 900 - 4 = 896. This implies that a computer program based on the use of
0 0 0 0 expanded element stiffness matrices would be highly inefficient, as it would perform
0
~
a sequence of operations with numbers equal to zero. The same arguments hold for
0 0 0 0 0 the expanded element load vector.
K••-
3 -
0 0
fH = (9.96) It appears that it would be much more efficient to establish the stiffness matrix
25 - 25 100
and the load vector for an element using only those degrees of freedom that are
0 0 -25 25 100 related to the specific element in question.
From (9.85), we obtain with (9.94 )-(9.96) When (9.83) was established for element a, use was made of the approximation
for the temperature T given by (9.4) and valid for the entire body. Obviously, as we
25 -25 0 0 100 only consider one element we may equally well consider the approximation for T
-25 50 -25 0 200 applicable for this specific element only. This approximation is given by (7.127), i.e.
K= fl =
0 - 25 50 -25 (9.97)
200
(9.98)
0 0 -25 25 100
~s ex~ected, K and f1 established in this manner correspond exactly to the expressions where N• is the element shape function matrix and a• contains the temperatures at
giVen m Example 1 (cf. (9.2~) and (9..31)). Again we note the analogy of the assembling the nodal points of the element. Thus
process used above to that discussed m Chapter 3 (see for instance (3.23) and (3.26)).

N• = [N~ N2 ... (9.99)

9.3 FE formulation of one element
We have seen t~at the FE equation for the entire body can be established as the sum where n. denotes the number of nodal points for the element. From (9.98) we obtain
of the FE equ~t10ns for each element. We also noted that this summation was made
only f?r the stiffness matrix K and the load vector fl> since it is more convenient to
establish the boundary vector fb directly from (9.24 ).
B where
[3U N•
B• = --
dx
(9.100)
186 Introduction to the finite element method FE formulation of 1-D heat flow 187

T = Na K
i.e.

B• = [dN~ dN~ . .. dN~.J 0

/ 0
I 0
g lobal

(9.101) formu l ation

dx dx dx
L
Use of (9.100) in the weak form (9.3) and adoption of the Galerkin method evidently
result in

(J ~
s •T AkB• dx )a·= - [N•T AqJt. +I
~
N•TQ dx (9.102) expanded element

formul a lion
where, as before, La is the region of the element ex considered and where the notations
L a.
(9.78) and (9.79) have been utilized. If we define the foJlowing matrices:
I·
K• = f L,
B•T AkB• dx
element
0
f~ = -
0
[N•T AqJL_ (9.103) formula l i on

t; = f
L,
N•TQ dx I·
Figure 9.14
·I
The three different FE formulations

then (9.102) can be written as locol numbers: 2

f~ + Cf
0 § § 0 §
Q
I x•a• = (9.104) global numbers: 2 3 5 6 7 8 9
~

x• is the element stiffness matrix for element ex, f~ is the element boundary vector for \
element con sidered
element ex and ff is the element load vector for element ex. Moreover, if we define the
c·
element force vector for element ex as Figure 9.15 Local and global nodal point numbering

r·= fb+ ff 1 (9.105)

formulation, one element is considered, but the global approximation T = Na is still
then (9.104) may be written as used. Finally, in the element formulation, one element is considered and the
approximation T = N•a•, applicable to this particular element, is used.
K• a• = f• (9.106) Therefore to obtain an efficient FE computer program the element stiffness
' .
matrix K• is derived for each element. Then, instead of using the inefficient formulatiOn
This system of equations constitutes the FE formulation of one element. Contrary (9.85), each term of K• is added directly to the proper component of the (globa~)
to its expanded form, (9.106) contains only those degrees of freedom which belong stiffness matrix K. In a similar manner, each term of the element load vector ff IS
to the element considered and it folJows that the non-zero components of K•• and added directly to the proper component of the (global) load vector f, . To describe
t;• are those given by K• and ff, respectively. We recalJ that the dimensions of K• this procedure, we first note that in o rder to obtain a standardized determination of
and fj' are n. x n. and n. x 1, respectively. the FE equations (9.104) for an element, it is advantageous to use a local nodal point
The three different FE formulations - global formulation. expanded element numbering which should be distinguished from the global nodal point numbering.
formulation and element formulation - are illustrated in Figure 9.14. We recall that, To firm up these ideas, consider the body shown in Figure 9.15, where,
in the global formulation, the entire body is considered and the approximation for convenience, simple linear elements are used. The element given by the global
T = Na, applicable to the entire body, is adopted. In the expanded element numbers 5 and 6 is considered. Using the local numbering, the FE equations (9.104)
 FE formulation of 1-D heat flow 189
188 Introduction to the finite element method
local n oda l points
take the form
1
I \2
(9.107) ~----C===~~--- x
x= 2 =4
t>l.1
As local point number 1 corresponds to global point number 5 and local point
number 2 corresponds to global point number 6, we obtain the following procedure: Figure 9.17 Local nodal point numbering of element 1
5 6 K~ 1 is added to K 55
From (9.103) the element stiffness matrix K• of element l becomes
K~ 2

K21
isaddedto K 56
is added to K 65
K i 2 is added to K 66
(9.108)
K• = f: BeT AkB• dx (9.109)

According to (9.26) we have Ak = 50 and using the n •-matrix as given by (7.35), we

s[n1
6 fh
J= /i1
n2
is added to /. 5
is added to !.6
obtain
-1] dx
1
It appears that the identification of proper components ofK and f 1 as given by (9.108)
is purely a geometrical or topological scheme. In a general situation, the information (9.110)
on how local numbers communicate with global numbers is given by the so-called
topology data.
Let us summarize the discussion above. The FE equations can alwa-ys be i.e.
established directly from (9.16), (9.15) and (9.13). The stiffness matrix K and the load
vector f 1 may also be derived according to (9.85) using the expanded form of the FE K• = [ 25 -25] (9.111)
equations for the individual elements. However, in order to construct an efficient FE - 25 25 ~
computer program, the stiffness matrix K and the load vector (1 are established using The element load vector f j is given by (9.103) and as the load Q = 100 (cf. (9.26))
the element stiffness K• and element load vector f j given by (9.103), where the
we obtain for element 1
components of K• and fj are added to K and f 1 using the topology data. In any case,
the boundary vector fb is most conveniently obtained by using (9.13 ). r· =
I
f4 N•TQ
2
dx = g
L
f4 [-X+ 4]
2 x-2
dx = 100
2
f4 [-X+ 4]
2 x-2
dx (9.1 12)

9.3.1 Example 4 where (7.29) has been used. The result is that
(9.113)
We shall now show the manner in which the FE equations for the entire body can r• = [100]
I 100
be established using the element stiffness K• and element load vector fj given by
(9. 103). Again the problem discussed in Examples 1, 2 and 3 is considered. For According to Figures 9.16 and 9.17 the local nodal point 1 corresponds to the g~obal
convenience, the element numbering and global nodal point numbering are shown nodal point 1 and the local nodal point 2 corresponds to the global nodal pomt 2.
again in Figure 9.16. The contributions of (9.111) and (9.113) are added to the (global) stiffness ~atri~ K
We consider element 1 first with the local nodal points as shown in Figure 9.17. and the (global) load vector f 1 and as these are originally zero and of dtmensJOn
4 x 4 and 4 x 1, respectively, we obtain the intermediate result

25 -25 0 0 100
e l.1 el.2 el.3
2 3 I. - 25 25 0 0 100
(9.114)
X
K= rl =
x=2 =I. =6 =8 , 0 0 0 0 0

Figure 9. 16 Global nodal points and element numbering 0 0 0 0 0

 FE formulation of 1-D heat flow 191
190 Introduction to the finite element method

local nodal points

the following final result for the global stiffness matrix K and the global load vector f1:
I \ 25 - 25 0 0 100
1 2
~--------~====~----
x=l. =6
x - 25 50 ·-25 0 200
(9.116)
K= fl =
el 2 0 - 25 50 -25 200
Figure 9.18 Local nodal point numbering of element 2 0 0 -25 25 100

local nodal poi nts As expected, these K- and f 1-matrices correspond exactly to those obtained in
I
1
\2 Examples 1 and 2 (cf. (9.29), (9.31) and (9.97)), respectively. Finally, we recall that
the boundary vector fb is obtained as in Example 1, i.e. it is given by (9.30).
~-----------------E====~-----x
X: 6 :8
el 3
Figure 9.19 Local nodal point numbering of element 3 9.3.2 Use of local coordinate systems

We next consider element 2 with the local nodal points as shown in Figure 9.18. In order to standardize further the determination of the element stiffness matrix K•
From (9.103) the element stiffness matrix K• of element 2 is and the element load vector ff, we may use a convenient local coordinate system.
Since K • and ff are obtained by integration (cf. (9.103)), it is obvious that we may
K• = L6
BeT AkB• dx choose any coordinate transformation when these integrations are carried out.
To show this, we adopt the linear transformation
With (7.35) and as Ak = 50 we obtain ' x = c~ +d (9.1 17)

where c and d are parameters and ~ is the new variable. As shown in Figure 9.20,
K·=f6[-1 JL]Ak[ -_!_ _!_Jd.x let the coordinates of the left and right ends of the element be given by X; and x i ,
4 1/ L L L
respectively, i.e. the length L of the element becomes L =xi- X;.
and a comparison with (9.110) shows that the element stiffnesses K• for elements 1 Now we choose the coordinate transformation (9.117), for instance, such tha~
and 2 are identical. This is certainly not surprising, since Ak and L are identical for ~ = - 1 and ~ = 1 at the left end and right end of the element, respectively. That is,
the two elemen~s. Moreover, it can easily be shown that the element load vector ff in the local ~-coordinate system, the element of Figure 9.20 takes the form shown in
fo_r element 2 1s equal to that of element 1, i.e. it is given by (9.113). From Figure 9.21.
F1gures 9.16 and 9.18 the topology data of element 2 is that local nodal point 1
corresponds to global nodal point 2 and local nodal point 2 corresponds to global
nodal point 3. Therefore, adding (9. 111) and (9.113) to the proper components of
(9.114) results in poss1ble i nter1or nodal points
\ I
-25 0 0 '::3~>----X
25 100 0 0

X ·I
-25 50 - 25 0 200
K= fl = (9.115) L
0 - 25 25 0 100
Figure 9.20 Geometry of one element in global coordinate system
0 0 0 0 0
We then consider element 3 with the local nodal points as shown in Figure 9.19.
It can easily be shown that the element stiffness matrix K• and the element load possibl e inter tOr nodal points
v~ctor ff are again given by (9.111) and (9.113), respectively. According to
\ I 0/ ::::9----- ~
F1gures 9.16 and 9.19 the topology data for element 3 is that local point 1 corresponds -1 0 1
to global nodal point 3 and local nodal point 2 corresponds to global nodal ~int 4.
Therefore, adding (9.111) a nd (9.113) to the proper components of (9.115) gives Figure 9.21 Geometry of one element in local e-coordinate system
192 Introduction to the finite element method FE formulation of 1-D heal flow 193
This transformation is easily shown to be given by can be written as
L
2
1
x = -<+- (x- + x-)
2 I J (9.118) fj=
fx, N•TQ dx= -1
x, 2
f
1

- 1
[ 1
1+
<]e Q-L d<
2
(9.125)

which gives and if Q = constant we get

L
dx = - de (9.119)
2 (9.126)

As an example, consider the simple linear element of Chapter 7. From (7.28) and
(7.29) we have It appears from (9. 124) and (9.126) that for constant Ak- and Q-values the element
stiffness matrix K• and the element load vector fj can be established directly using
Nc = [Nf Nj] (9.120) closed-form expressions. In an FE program such expressions clearly facilitate
where computational efficiency and similar closed-form solutions may be established for a
sequence of typical forms of Ak and Q. Here we have only considered the simple
c
N;(x) = - -1 (x- xi)= - -1 [L- <+ -I (x + x -)- x .J= -1 (1- ;)
1 1 1
linear element, but similar closed-form expressions may be derived for higher-order
L L2 2 2 "' elements.
c1
Ni(x)=-(x-x;)=
L
-I - <+-(x
L2
I
2 1
[L
I 1+ ; )
+x .) - x . =-(
J I 2 "'
J (9.121)

Moreover, from (7.33) we have that

9.4 Axially loaded elastic bar
Consider the axially loaded elastic bar shown in Figure 9.22, where A(x ) is the
(9.122) cross-sectional area, E( x ) is Young's modulus and b(x) is the body force per unit
axial length.
With u being the displacement measured positive in the x-direction we have
in accordance with (7.35). From (9.1 J9) and (9.122) it follows that the element stiffness
matrix K• given by (9.103) is
(9.127)

K• = f x; B•T AkB• dx =
x,
J [ - ! /L]Ak[ - ~ ~J~d<
1

- t 1/ L L L 2
where a now contains the displacements at the nodal points. The strain e is given by

= _!_ fI Ak[ 1 - Jl J d< G-=-:l (9.128)

~
2L - t - 1
i.e.
In Chapter 4 it was shown that the equations for such a problem are similar to those
Kc = _!_[ 1
2L -1
-l]J1
1

- 1
Akd< (9.123)
of beat flow; cf. (4.8)- (4.10) with (4.18) and (4.20), (4.21 ). Referring to (9.12) and

E(x ) A (x)
If Ak = constant we get

1- _'\__,.~~ - j
(9.124) b(x)

' ~-----+------------------~------ x
a b
From (9.119) and (9.121 ) it follows that the element load vector fj .given by (9.103) Figure 9.22 Axially loaded elastic bar
194 Introduction to the finite element method
FE formulation of 1-D heat flow 195

(r BT AEB dx )a= [NT Acr]~ + r

Table 4.1 we can therefore write the FE formulation directly as

NTb dx (9.129)

where cr is the normal stress. Care should be taken not to confuse the axial load b
r·=bL[1]
I 2 1

where k is the stiffness of the bar (cf. (3.38)).

(9.136)

with the position x = bat the right end of the bar. The force N in the bar is given by For the heat flow problem, the global stiffness matrix K is singular ( cf. (9.38)),
whereas any submatrix K of K is non-singular ( cf. (9.49)). The introduction of the
du
N = Acr; cr = Ee = E - (9.130) boundary conditions by which at least one nodal temperature is specified implies
dx that K could be established from K. The present case of an axially loaded bar is quite
N is seen to be positive if a tensile stress exists in the bar. Let us define the following similar and the stiffness matrix K of (9.131) is also singular. The introduction of
matrices: boundary conditions in terms of the specification of at least one nodal displacement

K = r BTAEBdx
is now the tool which enables us to obtain a non-singular submatrix K from K. This
can be proved formally just as for heat flow, but the need to specify at least one
nodal displacement is physically obvious since it removes possible rigid-body motions.
We recall that a rigid-body motion does not creat~ any strains.
fb = [NT Acr]~ (9.131)

f1 = Lb NTb dx 9.4.1 Example S

With the discussion of the element stiffness matrix K• and the element load vector
as well as fi and the introduction of a convenient local coordinate system for the establishment
of these matrices, we are in a position to illustrate the details of the efficient
I f = fb + fl (9.132) implementation of the FE equations into a computer program. For this purpose
consider the axially loaded elastic bar shown in Figure 9.23. The bar consists of two
Then (9.129) can be written as parts, 1 and 2, each having a constant Young' s modulus E, constant cross-sectional
if'a A and constant external load b.
I Ka = f I (9.133) This problem is analyzed using the two simple linear elements shown in
Figure 9.24.
where it is easily seen that the force vector f has the dimension of force, i.e. [NJ
Next let us consider the FE formulation of one element. The element stiffness E1A1 EzA2
matrix K• and the element load vector fi now take the form I _.... _..
x =a \b
-:- --- -I
\

"'b 2 x=b
X

(9.134) I· L, I· Lz ·I
Figure 9.23 Configuration of axially loaded elastic bar

E1A1b 1 E 2 A2 b2
For constant AE- and b-values, these expression can, analogous to (9.124) and (9.126),
) )
be written as ·
, 2 3

K• = k[ -1l -1]·
1 '
k = AE
L
(9.135)
\ I·
L,
I·
L2
X

Figure 9.24 Two simple linear elements used to model the problem of Figure 9.23
196 Introduction to the finite element method FE formula ti on of 1-D hea t flow 197

Let us consider first the boundary vector fb given by (9.131). As in (9.24) we obtain It appears that N x=a is that tensile force - namely, the reaction - which must be
applied at x = a in order to maintain equilibrium of the bar.
- (Aa)x=aj
fb = 0 (9.137)
[
( AO" )x=b
Next we evaluate the element with length L 1 . From (9.135) and (9.136) we have 9.5 Basic features of an FE computer program ¥"/
K· = 1 -1]·k1 [
- 1 1 '
r· =~[1]
I 2 1
We are now in a position to sketch the computer flow diagram for an FE program
as follows:

where k 1 = A 1E 1/L 1. The topology data follows from Figure 9.24; that is, after t9-e
contributions for the element with length L 1 , K and f1 are given by Input data:
Number of equations n
k1 - k1 01 [b1L1l Number n.1and type of elements

K~ [ -~ ~ ~J r,~~ b,:,J Position of nodal points

Topology data
Loading data
Material data
For the element with length L 2 we obtain

kz[ - -1]· r·=bzLz[1] 1

K• = 1
1 1 ' I 2 1 I l K=O; r) = o 1

where k 2 = A 2 E2 / L 2 • With Figure 9.24, after the contributions of this element we 1

obtain Fori = 1 until i = n.1do
Calculate K• and f f for element i
add K• and f~ to proper positions of K and (1
(9.138)
l
Input data:
From (9.137) and (9.138) the final FE equations are Boundary conditions

kl -k) 1
- kl k1 + kz (9.139) I Calculate fb I
[
0 - k2
1
The boundary conditions are now introduced. We assume that the bar is fixed
at x = a and that the right end of the bar is free, i.e.
I Establish Ka = fb + (1
(9.140) 1
The solution becomes Modify system of equations according to the boundary conditions
and solve the modified system of equations

1
(9.142)
I Calculate and write results (temperatures, fluxes, etc.) I
198 Introduction to the finite element method FE formulation of 1-D heat flow 199

First a sequence of input data is read. Then the total stiffness matrix K and the total /A(x)
load vector f 1 are initialized, i.e. the matrix dimensions are identified and all their
components are set to zero. For each element, the element stiffness matrix K• and
,___...._p______]_J - q n =a ( T- T00 )

the element load vector ff are calculated and added to the proper components of K
and f 1 using the topology data. This is the assembling process. The boundary conditions
are read and the boundary vector fb is calculated. Those components of fb that are
unknown (i.e. the 'reactions') are included in the matrix r of (9.57). The entire system
LLWm=LLlJ Q (x)

0 b
of equations is established, and after the introduction of the boundary conditions, 1---+----- - - - - - - + - - - - x
the modified system of equations is solved according to (9.58). The component r of
f which was originally unknown is then calculated using (9.59), i.e. the 'reactions' Figure 9.25 Convection at the right end of the fin
can now also be calculated. Finally, the results in terms of temperature, fluxes and
'reactions' are calculated and written out.
In practice, a so-called pre-processor is often used in connection with the input
to the program. This pre-processor may perform an automatic division into elements fin as shown in Figure 9.25;/F'cording to (9.14) the FE formulation is
following some rules specified by the user, and it may also provide graphical
information on the element mesh, nodal points, etc. This facilitates the possibilities Ka = fb + f1 (9.144)
for ensuring that correct input data is used. Likewise, a so-called post-processor is The boundary vector fb given by (9.13) can be written as
often used to provide graphical information on the results, for instance in terms of
contour curves for the temperature etc. fb = - [NT Aq]~ = - (NT Aq)x=b +(NT Aq)x=a (9.145)
As we have one-dimensional heat flow, the flux qn of Figure 9.25 is qn = q(x = b),
i.e. (9.143) takes the following form of a boundary condition:

9.6 Heat flow with convection (9.146)

We note that if convection occurs at the left end of the fin, then qn = - q(x = a). In
Generally, fluid motion around a body is a very effective means of transporting heat general, we have the approximation T = Na, i.e. we may write the unknown
between the body and the fluid. This type of heat transport is called co~ection. temperature Tx=b as
Assume that the fluid far away from the body has the uniform temperature T~he
temperature at the surface of the body is given by T and it is natural to expect that
the temperature difference T - T00 controls the amount of heat flow between the This implies that (9.146) becomes
body and the fluid. This leads to Newton's convection boundary condition expressed
through (q)x =b = ctNx=ba - r:xToo
The boundary vector fb given by (9.145) can then be written as
qn=r:x(T-T00 ) I (9.143)
fb = -ct(ANT N)x=ba + r:xT00 (ANT)x =b +(NT Aq)x=a
where qn is the flux through the boundary with the outer unit normal vector n ( cf. Use of this expression in (9.144) gives
Figure 6.1 and (6.3)) and r:x is a parameter, the so-called convection coefficient with
the dimension [J/m 2 s oc]. We note the similarity of (9.143) and (6.31). The (K + Kc)a = r:xT00 (ANT)x=b +(NT Aq)x=a + f1
convection coefficient r:x depends on a variety of circumstances such as fluid velocity, (9.147)
where Kc = r:x(ANTN)x=b
type of fluid, surface conditions, etc. but for our present purpose we assume that the
convection coefficient r:x is known. Moreover, r:x is a positive quantity, implying that
when the temperature difference T- T00 is positive, the flux qn as given by (9.143) It may be of interest to evaluate the expression for Kc. The number of nodal
becomes positive. This is in accordance with the definition of the flux qn ( cf. points for the entire body is given by n. Assuming that nodal point n is located at
Figure 6.1 ), since a positive qn means that heat leaves the body. x = b we obtain
As an example of (9.143) assume that convection occurs at the right end of the Nx=b = [0 0 . . . 1]
200 Introduction to the finite element method FE formulation of 1-D heat flow 201
i.e. not truly one-dimensional. Actually, this situation is similar to the discussion related
to the beat supply Q, and referring to the discussion on page 48 we accept the
0 0 0 engineering approximation of a one-dimensional heat flow in the x-direction provided
0 0 0 that the temperature variations along the fin are much larger than terrperature
Kc = a(ANTN)x=b = aAb (9.148) variations normal to the fin. This assumption is usually realistic if the fin is thin.
With this discussion it is obvious that we may consider the heat flux q" due to
0 0 1 convection as a modification of the~t supply given by Q. Now Q is the heat supply
Expression (9.147) shows that the convection condition implies a modification per unit time and per unit length of the fin. Moreover, Q is positive when beat is
~f the stiffness matrix. The new stiffness matrix is symmetric just like K, but it is supplied to the fin. In contrast, q" is the heat per unit time and per unit area and q"
~mportant to observe that in contrast to K the modified stiffness matrix is non-singular, is positive when beat leaves the fin. Let P be the perimeter of the fin; that is, q"P is
I.e. the heat per unit time and per unit length of the fin. It appears that we may treat
the convection as a modification of Q according to the scheme
(9.149) (9.152)

To show this, consider the quadratic form I given by which means that Q should be replaced by the term Q - q"P. The FE formulation
is again given by (9.14), i.e.
I= a T(K + Kc)a =aTka+ a TKca (9.150)
Referring to (9.41) and (9.43) we have aTKa= 0 if and only if dTj dx = 0, i.e. if the
Ka = fb + f, (9.153)
components of a are equal; otherwise aTKa > 0. From (9.150) and (9.148) it therefore where the load vector r, is given by (9.13), i.e.
follows that
I > 0 for all a -=F 0 (9.151)
f1 = lb NTQdx (9.154)

!hat is, the modified stiffness matrix is positive definite and, according to (2.67), it
IS then ~on-singular. This proves that (9.149) is correct and implies that the system
Combining (9.152) and (9. 143) gives
of equatwns (9.147) can be solved directly irrespective of the boundary condition at Q ~Q - aPT + aPT<X)
the left end of the fin. "-
As ~nother important engineering problem of convection, assume that lateral The unknown temperature T can be replaced by T = Na to obtain
convectiOn occurs along the outer surface of the fin as shown in Figure 9.26. This Q·zr+Q- rxPNa + aPT 00
(9.155)
type o_f co~vec~ion may cause conceptual difficulties because it can only occur if heat
flows m duect10ns perpendicular to the x-axis and this means that the heat flow is With (9.153)-(9.155) we obtain

fTID:fiD1IlQ 0 =U(T-T00 )
(K + Kc)a = fb + f1+ r aPT00 NT dx
(9.156)
where Kc = lb aPNTN dx

..______,!-\
\ : - - - -_ _ ]
\A(x) and fb and f1 are given in the usual manner by (9.13). Again it appears that the
convection results in a modification of the stiffness matrix. The modified stiffness
LLLLllJniiJ
a b
a (x) matrix is symmetric, and in a manner similar to (9.150) and (9.151) it can easily be
proved that this matrix is non-singular, i.e.
r----+--------------------~------x
(9.157)
Figure 9.26 Lateral convection along the outer surface of the fin
202 Introduction to the finite element method FE formulation of 1-D heat flow 203
Hence the system of equations (9.156) can be solved directly irrespective of the ___.--- el emenl boundary
boundary conditions imposed on the ends of the fin. I
a) I
We finally observe that with the results above it is straightforward to combine "smoolhmg" ~
the convection cases of Figures 9.25 and 9.26.

~
N·I

9.7 C 0 -continuity b)
dN;
dx
We have already established criteria in Chapter 7 which ensure that the FE formulation
converges towards the exact solution for decreasing element size. According to page
93, the~e. _convergence criteria consist of the fulfilment of completeness and
compatJbJitty. The compatibility or conforming requirement stated that the
approximation of the temperature also varies in a continuous manner over element cl
boundaries.
Let us now derive this continuity requirement in a different manner. It is not
surprising that the fulfilment of the convergence criteria hinges on the expressions
chosen for the approximation, i.e. on the global shape function matrix N. Referring
to the glo~aJ stiffness matrix K, the boundary vector fb and the load vector f1 given
by (9.13 ), 1t appears that K places the most severe conditions on N since the derivative
B = dN f dx enters the expression forK. A component K IJ of K is given by (9.19), i.e.

Kii =
f
bdN-
a
-
dx
dN .
' Ak- ; dx
dx
(9.158)
'
- oo

If the term~ Ak and ~N;/dx are continuous functions, the integration given by (9.lfs) Figure 9.27 Differentiation of shape function with C 0 -continuity
can be earned out duectly. However, if discontinuities occur we may run into trouble.
~n order to evaluate this situation, assume that the global shape function N; varies Here we conclude that C 0 -continuity is sufficient for the global shape functions of
rn the manner shown in Figure 9.27.1t appears that N. is continuous whereas dN.fdx heat flow problems, but it is obvious that the continuity required depends on the
is_ disco_ntinuous over the element boundary. Moreo;er, even though dN;/dx v~ries differential equations involved. We shall return to this topic in later chapters.
dJscontrnuously, an integration which involves dNJ dx is still well defined.
To investigate whether it is also possible to make a well-defined integration of
d N;/dx , :'e repl~c~ N; by the smooth dashed curve shown in Figure 9.27(a). The
2 2

correspond~ng vanat1~ns of dNJ dx and d 2 NJdx 2 are shown in Figure 9.27(b) and 9.8 Concluding remarks
(c), respectively, and 1t follows that an integration involving d 2 N;/dx 2 is not,. in
general, well defined. We may draw the conclusion that in order for the stiffness In principle, even though this chapter has been devoted to simple physical problems,
mat~x to be est~blished in a well-defined manner, we are led to the compatibility it contains all the ingredients of an FE formulation. In later chapters more complicated
requ~rement statmg that the approximation of the temperature must vary in a physical problems will be treated, but the fundamental FE results are analogous to
contrnuous manner over the element boundaries. A function which is continuous but those discussed here. It is for this reason that we have presented a rather detailed
which bas a discontinuous first derivative is termed a C 0 -continuous function. In discussion of various typical issues related to the FE formulation. As these matters
general, we have the following statement: are easily generalized, we will adopt in later chapters a much more compact method
of presentation for establishing the different FE formulations.
C"-continuity means that the n-derivative is continuous. Let us therefore review some of the fundamental facts. The FE formulation can
always be given in the form of(9.16), (9.15) and (9.13) and this is typically the format
204 Introduction to the finite element method FE formulation of 1-D heat flow 205

method and this fact is illustrated in Figure 9.28. We also note that,~addition to
being more general, the Galerkin method often provides a more straightforward
derivation of the FE equations than the Rayleigh- Ritz method.
We finally draw the reader's attention to the comprehensive textbooks by Becker
et al. ( 1981 ), Hughes ( 1987) and Stasa ( 1985) that also treat one-dimensional heat flow.

Figure 9.28 Relation between the Rayleigh- Ritz method and the Galerkin method

chosen when discussing theoretical aspects. Alternatively, the global stiffness matrix
K and the global load vector f 1 may be derived as the sum of the expanded element
stiffness matrix K•• and the expanded element load vector q•, respectively, for all the
individual elements (cf. (9.85)). Conceptually (9.85) is of fundamental importance,
but when establishing a computer program for the FE method, these expanded forms
are extremely inconvenient to work with, so, instead, the element stiffness matrix K•
and the element load vector ff are derived using only those degrees of freedom related
to each individual element (cf. (9.103)). The contributions ofK• and ff are then added
to the global stiffness matrix K and the global load vector f 1, respectively, u~ng the
topology data. We also note that the boundary vector fb is always derived uyng the
form given by (9.13).
It appears that the FE method is extremely well suited for efficient implementation
on a computer, which enables us to deal with a variety of problems.
As a global shape function Ni only differs from zero in those elements which
contain the nodal point i, the global stiffness matrix K was found to be banded.
Moreover, the Galerkin method implies that K is symmetric. We found that K is
singular and positive semi-definite; see (9.38), (9.41) and (9.43). We proved that any
submatrix ~ obtained from K by deleting one or more rows and the corresponding
columns is symmetric, non-singular and positive definite; see (9.45), (9.48) and (9.49).
These aspects have important consequences when the boundary conditions are
prescribed. We also showed that, apart from problems involving convection, the
specification of at least one nodal temperature is a necessity in order to obtain a
unique solution of the FE equations. Moreover, the balance equation for the entire
body is fulfilled through relation (9.68).
In this textbook, the Galerkin method is used universally, but we remark that,
as an alternative, one may also adopt the so-called Rayleigh-Ritz method. This
method relies on the construction of certain functionals for which a minimum is
sought using the calculus of variations. If such a functional can be established, it turns
out that the Rayleigh- Ritz and Galerkin methods result in identical FE formulations.
For a discussion of these topics the reader is referred to Hughes (1987), Strang and
Fix (1973) and Zienkiewicz and Taylor (1989). However, the important point is that
whereas the Rayleigh- Ritz method can be used only for certain types of differential
equations, the Galerkin method is applicable to arbitrary differential equations. This
means that the Rayleigh- Ritz method can be viewed as a subset of the Gajerkin
 207
10 _ _ _ _ _____ y
FE formulation of 2-D and 3-D heat flow
....
FE formulation of two- and
three-dimensional heat flow

Figure 10. 1 Two-dimensional region A with boundary fL' = !L'h + ff',

Moreover, the flux q. at the boundary of the region is gi ven by Figure 6.1 and (6.3), i.e.
Following the detailed discussion of the FE formulation for one-dimensional heat ( 10.5)
flow, it is an easy task to treat two- and three-dimensional heat flow. We emphasize
tha t even though we focus on heat flow here, a sequence of other important physical and qn is positive when heat leaves the body. As shown in Figure 10.1, the boundary
problems is governed by similar equations (see Table 6.1 ). Therefore, the FE Sf of the region A consists of 2, and ~· On 2, we have q. = h where h is a known
formulation presented in this chapter has significance in a variety of other fiJds. quantity and on ~ we have T = g, whereg is a known quantity. With these preliminary
remarks we are in a position to obtain the FE formulation.
The temperature Tis approximated by means of (7.137), i.e.
T = Na (10.6)
where N is the global shape function matrix and a contains the temperatures at the
10.1 Two-dimensional heat flow nodal points in the entire body. This means that
'
Considering two-dimensional heat flow first, the strong form was given by
( 6.32 )- ( 6.34) and the corresponding weak form by ( 10.7)

(Vv)TtDVTdA =- f vhtd2 - f vq.td!t' +J vQtdA (10.1)

fA !e, ~ A
where n is the number of nodal points for the entire body and a component N;
T =g on ~ ( 10.2) depends on x a nd y, i.e. N ; = N ;(x , y). From (10.6) we obtain
(cf. (6.46) and (6.47)). In (10.1) and (10.2) we recall that v is the arbitrary weight
VT = Ba where B = VN (10.8 )
function, D the constitutive matrix, T the temperature, t the thickness and Q the heat
supply per unit time and p.er unit volume of the body. The flux vector q is dete~ined which implies that
from Fourier's law ( 6.4 ), i.e.
q = -DVT ( 10.3) ... aa:·] ( 10.9)
where the temperature gradient VT is aN.
ay
in accordance with (7.140} and (7.142). Inserting (lO.f) into (10.1) gives
( 10.4)
(Vv)TDBt dA)a = - f vht d.sf- f vq.t d2 + f vQt dA ( 10.10)
(fA !e, ~ to A

206
208 Introduction to the finite element method FE form ulation of 2-D and 3-D heat flow 209
The final step is to choose the arbitrary weight function v. In accordance with i.e. ( 10.16) becomes
the Galerkin method we set
v=Nc (10.11) I Ka =f I ( 10.18)

Since v is arbitrary, the matrix c is arbitrary. From ( 10.11) we obtain Just like one-dimensional heat flow, the force vector f has the dimension of [J/ s]
and a comparison with the corresponding formulation for one-dimensional beat flow
Vv = Be ( 10.12)
(cf. (9.13)- (9.16)) shows that a complete similarity exists.
As v = vT, ( 10.11 ) can also be written as Owing to the symmetry of the constitutive matrix D (cf. (6.10)) it follows from
= cTNT
( 10.15) that K is symmetric "-
V ( 10.13)
Inserting ( 10.12) and ( 10.13) into ( 10.10), and noting that cis independent of position, (10.19)
gives

cT[ (t BTDBt dA )a+ J 2t NTht d.s!' + f~NTqnt d.5f- L NTQt ~0

Similar to the evaluation of (9.38)- (9.40) it follows tha t K is singula r, i.e.

I det K = 0 I ( 10.20)
As this expression should hold for arbitrary cT-matrices, we conclude that
Moreover, since D is positive definite (cf. (6.8)), we derive in a similar manner to

(LBTDBtdA )a=- J..s!l.~ht d.sf- f~NTqntd.5f+ LNTQtdA (10.1 4)

(9.41)- (9.43) that K is positive semi-definite, i.e.

(10.21)
which is the FE formulation sought.
To write (10.14) in a more compact fashion, we define the following matrices: for all a # 0 and that only if the temperature gradient is zero, i.e. VT = Ba = 0, is
the quadratic form above equal to zero. Finally, using the approach discussed in

K= L BTDBtdA
relation to (9.44)- (9.49) it can easily be shown that any submatrix Kobtained from
K by deleting one or more rows and the corresponding columns is symmetric,
non-singular and positive definite. With reference to Chapter 9, we conclude that at

fb = -J 2,
NThtd.5f - f.
~
~qntd.5f ( 10.15)
least one nodal temperature has to be prescribed in order to obtain a unique solution
of the FE equations. A systematic consideration of the boundary conditions is again
given by the approach outlined in (9.57)- (9.59).
Let us now prove that the components of the force vector again fulfil relation

As D has the dimension 2 x 2 and B the dimension 2 x n, it follows that K is a

L Qt dA = Pz q"t,d.sf
-
(9.68). As we consider a region A with thickness t, the balance principle states that

( 10.22)
square matrix with dimension n x n and it is the stiffn ess matrix. Likewise, both fb
(cf. (6.19)). Let us write (10.17) as
and f 1 have the dimension n x 1 and they are termed the boundary vector and load
vector, respectively. With (10.15), ( 10.14) can be written as /; = fbi + /r;; i = 1, . .. , n
which leads to •
I Ka = fb + r. (10.16) n

I.~; =
n

2: fbi + 2: /r;
n
(10.23)
i= 1 i= 1 i= 1
We define the force vector f by According to (10.1 5) we have that

( 10.17) fbi = -J 2,
N;ht d.sf- f..21
N;q"t d.5f (10.24)
210 Introduction to the finite element method FE formulati on of 2-D and 3-D heat flow 211
We recaH that the boundary conditions specify the flux qn = h along.?,, whereas the For such a line source, the component};; of the load vector given by ( 10.26) becomes
flux qn along ~ is unspecified beforehand. Therefore, ( 10.24) may be rewritten as

fbi= -f.5!" Niqnt d2 (10.25)

};; =I N;(X, y)Qt dA
A
= rb· fa+N;(X, y)Q,b(x- a)b(y -
J
b- a-
b)t dx dy

From (10.15) a component of the load vector is given by = N;(a, b)t(a, b)Q, (10.32)

If the position (a, b) of the line source coincides with the position of nodal point i,
fli = LN;QtdA (10.26)
then N;(a, b)= N;(X;, Y;) = 1, i.e. we obtain I
Using (10.25) and (10.26) in (10.23) leads to J;; = Q,t(x;, Y;) if the line source is located at nodal point i (10.33)

.± /;
t=l
= _A:J.se (.± N;)qnt d2 +I (.± N;)Qt dA
t=l A t= l
As in the one-dimensional case of Chapter 9, it follows that any distributed load Q
may be replaced by line sources located at the nodal points.
which with (7.139) results in The global stiffness matrix K and the global load vector f 1 given by ( 10.15) are
obtained b y integration over the entire region A. These integrations may be obtained
.±/;= _A:J.se qntd2 + I
•=I A
QtdA (10.27) as a summation of integrations over each element. In this way we are led to the
expanded element stiffness matrix K•• for element a and the expanded element load
A comparison with ( 10.22) shows that vector~· for element a given by

I .t.~. ~ 0 I (10.28) Kee = I A.

BTDBt dA
( 10.34)
in accordance with (9.68). This means that the balance principle for the body is
expressed by the fact that the sum of the components of the force vector f is equal
r.· = I NTQt dA
A.
to zero. We emphasize that (10.28) holds exactly even though the FE method is an
approximate approach. where A., is the region of element a. Moreover, it follows directly that
The components of the load vector f 1 are given by ( 10.26). Let us assume now
that the load Q is given in terms of a line source, where the heat supply is concentrated
to a point in the xy-plane and at this point a heat supply Q, per unit time and per
a= I
unit thickness of the region is prescribed. Q, is the strength of the line source and Pfcl
(10.35)
has the dimension [J/ s m]. In such a situation, we may express Q using Dirac's delta r. = I r.:
function in a similar way to (9.70)- (9.72), i.e. a=l

'
Q = Q,b(x- a)b(y- b)= { 00 if (x, y~ =(a, b) (10.29) where ne1 denotes the total number of elements and where K!• and ff: refer to the
0 otherWise pertineJ?.t quantities for element a. Conceptl!ally, relations ( 10.34) and ( 10.35) are of
where (a, b) is the position of the line source. By definition, we have that fundamental importance, but just as for the one-dimensional FE formulation these
relations lead to a very inefficient computer implementatio n. In order to identify the
f~oo J:oo Q dx dy = f~oo f~oo Q, b(x- a)b(y- b) dx dy = Q, ( 10.30) non-zero components of K•• and ff• directly, we consider the FE formulation for an
element using only those degrees of freedom which belong to that specific element.
As a result of ( 10.29), (10.30) may also be written as The a pproximation of the temperature over each element is given by
T = N•a• ( 10.36)
J:_+ J:_+ Q, b(x- a)b(y- b) dx dy = Q. ( 10.31)
where N• is the element shape function matrix and a• contains the temperatures at
212 Introduction to the finite element method FE formulation of 2-D and 3-D heat flow 213

the nodal points of the element, i.e.

~]
/ I 1 I I
I I 1
N• = [N~ N•2 · · · N•...] ·' a• = [ • ( 10.37) I I
--- t--- t--- 1--- global
I
1
A I
I
T""
where n. denotes the number of nodal points for the element. From ( 10.36) we obtain
---+ --t- --.!._ __ formulot•on
I I I
I I I
VT = B•a• where s• = VN• ( 10.38) I I

i.e.
r-----------,

[
oN~
oNi ... oNe,..] I I
• = ox ox ox I I
8 (10.39)
oN• oN• oN• I I expanded element

~
__1 __2 ... ~

oy ay oy I Aa. I
I I for mulat1on
I
A comparison with ( 10.14 )-( 10.18) shows directly that the FE formulation fo r one
element is given by
1 T= Na K ee
L_ ____ __ ____ j
l
I

I K•a• = re (10.40)
element
where the element force vector re is given by
•
I re = ~+~ I (10.41) formulation

Moreover, the element stiffness matrix K•, the element boundary vector ~ and the
Figure 10.2 The three different FE formulations
element load vector ~ for element a are given by

K• = f A.
B•TDB•t dA approximation T = Na, applicable to the entire body, is adopted. In the expanded
element formulation, one element is considered, but the global approximation T = Na
is used. Finally, in the element formulation, one element is considered and the
( 10.42)
approximation T = N•a•, applicable ro this particular element, is used.
In a computer program, the element stj,ffness K• and the "e lement load vector ~
ff = f A.
N·TQt dA are determined by ( 10.42) and the contributions to the global stiffness matrix K and
the global load vector f 1 are obtained using the topology data. The boundary vector
fb is always derived using the form given by (10.15). It appears that the approach
where A .. is the region of element a, !lha: is that part of the element boundary on described is completely similar to that discussed for the FE formulation of
which the flux q" = h is known, whereas 2ga: is the remaining part of the element one-dimerJ.sional heat flow. We also observe that just as for one-dimensional heat
boundary. We recall that the boundary conditions along an element boundary are flow, the stiffness matrix K given by (10.15) contains first derivatives of the global
unknown, except when an element boundary coincides with the boundary of the body. shape functions, i.e. we again require that these shape functions fulfil C 0 -continuity.
Similar to- Figure 9.14, the three different FE formulations - global formulation, When the nodal temperature a has been determined by (10.18), the temperature
expanded element formulation and element formulation - are illustrated in Figure Tat an arbitrary point in an element is given by ( 10.36) and the temperature gradient
10.2. We recall that in the global formulation the entire body is considered and the VT is given by ( 10.38). From the temperature gradient, the flux vector q at any place
214 Introduction to the finite element method FE formulation of 2-D and 3-D heat flow 215

in the body is obtained from Fourier's law ( 10.3). Therefore, when a has been y y
ol /qn= h =30
b)
determined from (10.18) all quantities of interest can be derived.
3
D l J t ' t c 4
( 2 ' 1)
k =4
10.1.1 Example Q =45 _..--T = 10
qn=h=O-
I= 1
X X
Consider the square panel shown in Figure 10.3, where the boundary along they-axis ~ \.qn =h=O 8 2
is insulated (i.e. q. = h = 0) and a constant flux q. = h = 30 J / m 2 s is prescribed
along the boundaries y = 1 m and y = - 1m. A constant temperature T = 10 oc is Figure 10.4 (a) Problem reformulation; (b) finite element mesh
prescribed along x = 2m and a constant heat supply Q = 45 J j m 2 s is applied all
over the panel. The thickness tis constant and is equal to 1m. The material is assumed
to be isotropic and homogeneous, i.e. according to (6.13) and (6.14) the consti~e
matrix D can be written as
h

D = k[~ ~] = kl (10.43)

where the thermal conductivity is given by k = 4 J rc m s.

It is obvious that the heat flow in the panel is symmetric about the x-axis, i.e.
..
no heat flows across the x-axis. We may therefore reformulate the problem as shown
in Figure 10.4(a).
In order to solve this problem by the FE method using hand calculations only, Figure 10.5 Body with triangular base and height h
we adopt the extremely simple FE model consisting of two three-node triangles as
shown in Figure 10.4(b ). The element numbers and the global nodal numbers are
also indicated on this figure.
triangle (cf. (7.99)), we obtain
The element stiffness matrix K• and the element load vector ff are given by
( 10.42 ). With (1 0.43) and noting that B• is a constant matrix for the three-node K• = kB•TB•tAa ( 10.44)
where

B• ·= _1_ [yj- Yk Yk- Y; Y; - Yj] ( 10.45)

y
2Aa xk - xj X;-Xk xj-xi

As both the load Q and the thickness t are constants, ( 10.42) yields

...--T= 10 ~= Qt I A.
NeT dA • ( 10.46)

insulated 1----x 2
The element shape function matrix N• is given by (7.92) and ( 10.46) can be derived
using standard rules for area integration. In the present case, however, the shape
functions vary linearly, which means that ( 10.46) may be evaluated in a very simple
manner. For this purpose consider the body shown in Figure 10.5. The volume V of
this body is known to be given by ,-'

2 V=thA do.47)
where his the height and A the area of the triangle which forms the base of the body.
Figure 10.3 Square panel with heat flow Reference to Figure 7.23 shows that the element shape functions vary in the
 216 Introduction to the h"nite element method FE formulation of 2-D and 3-D heat flow 217

y Next we consider element 2 of Figure 10.6(b) where i = l,j = 2 and k = 3, i.e.

a) b) y
3 xi = Yi = 0; x1 = 2, y1 = 1; xk = 0, Yk = 1
( 2 .1} o - I
The area A 2 is A 2 = 1 and B• as given I?{ (10.45 ) becomes
Be = -
1[ 0 1 - 1 J-G
(
~
0 '::
t J.j
r:
0
I -
'1
( 10.51)
2 2 -2 0 2 ;t. - ~ t:
- J
Figure 10.6 Local nodal points for element I and 2 From ( 10.44) and (10.48) the results for the second element are then given by

manner given by Figure 10.5, i.e. from (10.46) we obtain with (10.47)
K• =[ ~ 0-4] 1 -1 ; ff = 15 [1] I

~~!A·[]
- 4 -1 5 1
( 10.48)
F, Comparing Figures 10.4(b) and 10.6(b ), the topology data for this element is th~
the local nodal points 1, 2, 3 correspond to the global nodal points 1, 3, 4 respectively.
This result is certainly not surprising as it shows that a constant load Q contributes For clarity, and using this topology data, the expanded element stiffness matrix Kec
equal amounts to each nodal point. and the expanded element load vector ff• are
The local nodal points of the two elements are shown in Figure 10.6 and in 1
4 0 0 -4
accordance with Chapter 7 these nodal points are listed in the counter-clockwise
direction 1, 2, 3. We first consider element 1 where i = 1, j = 2 and k = 3, i.e. 0 0 0 0 0
Kcc- Cf~ = 15
2 -
~ .!, 3 X;= Y; = 0; x1 = 2, y1 = 0; xk = 2, Yk =1 0 0 -1 1

With these values and as the area A 1 = 1, Be as given by ( 10.45) becomes -4 0 - 1 5

- I 0lB• = ~ [ - 1 , . 1 OJ _ 0 - I f-'t3 0 Adding the~e results to ( 10.50) we then obtain

I -:l 2 0 -2 2 -: 1-iO 1-'" _v (10.49)
• J
5 -1 0 - 4 2
C'Fro1n (10.44) and (10.48) and as the thickness t = 1, the results for the first element
are given by 1' - • t.t .. - 1 5 -4 0
K= r. = 15 • (10.52)
-1 0 -4 5 - 1 2

- 4 0 -1 5 1
5
-\ 0
- 4 Let us now evaluate the boundary vector fb given by ( 10.15). From Figure 10.4
we conclude that
Comparing Figures 10.4(b) and 10.6(a), local nodal points 1, 2, 3t..torrlfspond to the
global nodal points 1, 2, 3 respectively. That is, after the contributions from the first
element, the global stiffness matrix K and the global load vector f 1 are given by

1 - 1 0 0 i.e.
-1 5 - 4 0 (10.53)
K = Kf• =
0 -4 4 0
r. = m= 15 1
(10.50)

0 0 0 0 0 The only shape functions which differ from zero along the boundary DC a re N 3 and
218 introduction to the finite element method
FE form ulation of 2-D and 3-D heat flow 219
N 4 . Moreover, as they vary linearly a nd ash = 30 along DC we have Use of this solution in the second and third row of ( 10.57) implies that

f N Th d.sf = 30 f
0
0
d.sf = 30
0
0
( 10.54)
f .sl'ec
N2 qnd..Sf=25; f .sl'ec
N3qnd..Sf = 5 ( 10.59)

.sl'oc .sl'oc N3 As N 2 and N 3 are positive functions, (10.59) implies that q" is positive along the
N4 1 boundary BC, i.e. we conclude that in order to maintain the temperature a t T = 10
Likewise, it follows that along the boundary BC, heat must be extracted from the body along this boundary.
With the result of ( 10.59) it follows from ( 10.57) that the balance principle for the
0 body is satisfied in accordance with (10.28).
Let us finally determine the flux vector q in each element. From ( I0.3) and ( 10.3!J'1
we get
( 10.55) q = -DVT = - kVT = -kB•a•
where k = 4. For element 1 we obtain with ( 10.49) that
1
- 1 1
Using ( 10.53 )- (I 0.55) and as the thickness t = 1, the boundary vector fb is given by q =-4 X t[ 0
-2
0

f
whereas for element 2, where ( 10.51) holds, we get
N2qn d ..Sf
.sl'ac

30 + f.s!Bc
N3qn d..Sf
(10.56)
Q= - 4 X t[ O
-2
I
0
•
30
The FE equations for the body are obtained from ( 10.52) and (10.56), i.e. 10.1.2 Two-dimensi onal heat flo w with convection
0
P reviously, we assumed that the boundary ..Sf consists of the two parts, .!l',. and .Pg,

- 1
5 - 1
5 -4
0 - 4
0
Tt
10
f .sl'ec
N2qnd..Sf
+ 15
2 where the flux q" and the temperature Tare prescribed, respectively. Let us now also
assume that convection occurs on part of the boundary and let ~ denote that part.

- 4
0 - 4
0 - 1
5 - 1
5
10
T4
30 + f N3qn d ..Sf
.sl'oc
2
(10.57) Referring to (10.15) this affects only the boundary vector fb, which now becomes

fb = - f NTht d..Sf- f NTq"t d..Sf- f NTq"t d ..Sf ( 10.60)

30 ~ ~ .sl'c

where T2 = T3 = 10. From the first and fourth row we obtain where .!l;., ~ and ~ now comprise the entire boundary, as shown in Figure 10.7.
The convection along ~ is again given by Newton's convection boundary condition

[-~ -:J[i] = [;~ =3~ : ~~] = [ ~~] (9.143), i.e.

q" = a(T- T..,) ( 10.61 )

with the solution
To consider the influence of convection, the procedure discussed in Chapter 9 is
followed, i.e. the unknown temperature Tin ( I0.61)is replaced by T = Na. This gives
( 10.58)
q" = aNa - aT..,
220 Introduction to the finite element method FE formulation of 2-D and 3-D heat flow 221

where V is the volume of the body and S is the boundary surface which consists of
Shand S , where the flux qn = hand the temperature T = g are prescribed, respectively.
9
The temperatu re is again approximated by means of
( 10.65)
T = Na
q n =h where

( 10.66)
•
Figure 10.7 Two-dimensional region with boundary !L' = !i'h + !L', + !L'c
and n is the number of nodal points for the entire body and a component Ni now
and use of this expression in ( 10.60) results in depends on x, y and z, i.e. Ni = Ni(x, y, z). From (10.65) we obtain

fb = -f NTht d£'-
..'ll.
J NTq"t d£'- (f cxNTNt d£')a + J NTcxt
.Sf. .!Eo
T00
.!Eo
d2 i.e .
VT = Ba where B = V'N ( 10.67)

With this expression, the FE formulation of (10.16) takes the form oNl aN 2 aN.
ax ax OX
(K + Kc)a = -I !4
NTht d£'- I ~
N'fqnt dS!' + T00 J NTcxt d£' + f
~
1
(10.62)
oN 1
B= -
oy
oN2
-
ay
oN.
oy
•
( 10.68)

where Kc =I
.!EoaNTNt d£'
aN,
az
oN2
az
oN"
oz
The derivation of the FE formulation from ( 10.64) using the approximation
It appears that a modification of the stiffness matrix occurs. This new stiffness matrix
above in combination with the Galerkin method follows the same procedure as for
is symmetric, just like K, and similar to the discussion of (9.149) it follows that the
modified stiffness matrix is non-singular, i.e. two-dimensional heat flow. We may therefore write the result directly as

I det( K + Kc) # 0 I (10.63) I Ka = f I (10.69)

We finally mention that we are able to consider convection occurring along the where
lateral surfaces of the region A, i.e. convection normal to the xy-plane. This situation ( 10.70)
can be dealt with by a modification of the load Q which leads to a modification of
the load vector f1 as in the procedure for one-dimensional heat flow (cf. (9.152 ),
(9.154) and (9.155)). This modification is straightforward and is left as an exercise and
for the interested reader.

10.2 Three-dimensional heat flow

K = t BTDBdV

The weak form of three-dimensional heat flow is given by (6.49) and (6.50), i.e. fb = -I NTh dS - I NTqn dS
(10.71)

Iv
(Vv)TDVT dV =
s.
-I vh dS - I s,
vqn dS + r vQ dV
Jv (10.64)
s. s,

T = g on S9
222 Introduction to the hnite element method

Just like one- and two-dimensional heat flow, the force vector f has the dimension
[J / s]. Moreover, it is evident that the discussion for two-dimensional heat flow also
11
carries over directly to that of three-dimensional heat flow, and we simply mention
that convection now can occur only through a part Sc of the boundary surface. The
Guidelines for element
corresponding modification of the boundary vector fb is similar to that performed
for two-dimensional heat flow.
meshes and global nodal
We have seen that once the FE formulation for one-dimensional heat flow is
comprehended, it is a straightforward task to generalize these results to two- and
numbering
three-dimensional heat flow. It follows that we are now in a position to treat heat
flow for arbitrary geometries and with arbitrary material properties. By now it should
also be obvious why the FE method provides a unique tool for engineers and physicists.
We mention finally that two- and three-dimensional heat flow is also treated in the
textbooks by Bathe ( 1982), Becker et al. ( 1981 ), Hughes ( 1987), Johnson ( 1987), Stasa
(1985) and Zienkiewicz and Taylor (1989). We shall now discuss some general guidelines for the establishment of FE meshes
and for choosing the numbering of global nodal points. Also, the equation solution
process will be investigated in more detail.

11.1 FE mesh
The first step in an FE analysis is to select the type of elements and the corresponding
FE mesh. There are no fixed rules on how to make these decisions. Clearly, for a
given type of element the accuracy increases with decreasing element size and, in
general, one will use small elements in regions where the unknown function - the
temperature, say - varies rapidly. This means that a sound physical understanding
of the problem considered is of fundamental importance for a realistic analysis.
However, the decision on element types and size is more delicate than that. Every
analysis involves the use of resources, whether they are measured in terms of money
or manpower, and even though we aim at an accurate analysis, we do not want it
to be more accurate than required. For some problems, we are interested in detailed
information on the behaviour even in local regions, while for others we only want
to obtain a rather general and crude indication of the overall response. Moreover,
for some problems, simplifications may have already been introduced when the
problem was defined. For instance, we may have only vague ideas about the loading,
the boundary conditions and the material data. As engineers we must therefore use
our judgement in order to obtain that optimum choice for element type and element
mesh which balances the requirement of reliable results with that of cost effectiveness.
It is not surprising, therefore, that such a choice presents a delicate problem, and we
may refer to Cook et a/. (1989) and Zienkiewicz and Taylor ( 1989) for further
information. One often starts with a simple FE model, and these results may form
the basis for the establishment of a more refined model.
All finite elements are based on rather simple polynomial interpolations of the
unknown function within the element. For a given type of element this means that

223
224 Introd uction to the finite element method Element meshes and global nodal numbering 225

a) ~ surface b)

Figure 11.1 (a) Quadrilateral; (b) inferior division; (c) desirable division ins u lated

Figure 11.3 (a) Heat flow problem; (b) use of symmetry

> >- >---

~ >--- >---
/
a}
surface """ A E c b) A
I
'a
c
I
c)
E
H
c
'

~----~' I
I
Q
~ >- >- ..............
I "·
I I
l/
Figure 11.2 Examples of mesh refinement
L
Hp7777:
I
I
~' 2 1
P71
I

insu la ti on B F D B D F D
the smaller the elements, the greater the accuracy. However, it also implies that any (a) Heat flow problem; (b) use of symmetry ; (c) use of symmetry
Figure 11.4
dimension of an element should be kept as small as possible, i.e. it is not only the
size, but also the form of the element which is of importance. As an example, assume
that the quadrilateral shown in Figure 11.1 (a) is to be divided into two triangular provides the model of Figure 11.4(b). We also note that the flux q. is zero along the
elements. plane EF, which leads to the even simpler model of Figure 11.4(c).
It is obvious that the division in Figure 11.1 (c) is better than that of 11.1 (b),
since the largest dimension of the elements in I 1.1 (c) is smaller than that given by
11.1 (b). The ratio between the largest and smallest dimension of an element is called
the aspect ratio and in a good FE mesh, the aspect ratio is as close as possible to unity.
In order to obtain an efficient solution scheme, we want to use few elements in 11.2 Methods of solution
regions where the unknown function varies slowly, but many elements in regions
where it varies rapidly. Two possibilities, which allow for such a mesh refinement 11.2.1 Bandwidth
and which fulfil the continuity requirement, are illustrated in Figure 11.2 for the
three-node triangular element and four-node rectangle. We have seen that the FE method results in a symmetric global stiffness matrix K
To increase computational efficiency, symmetry properties should be used which is banded, i.e. with non-zero components clustered about the diagonal of K
whenever possible. An example was given in Chapter 10 on page 214. We recall that and zero components far away from the diagonal. With a cross indicating non-zero
symmetry involves not only geometry, but also loading and material data. As another components, this situation is illustrated in Figure 11.5 where all non-zero components
example of symmetry, consider a heating cable placed in the isotropic body shown fall between two lines parallel to the diagonal.
in Figure 11.3 (a). It is clearly advantageous to utilize the symmetry about the centre Including the diagonal, the distance from the diagonal to one of these lines is
line (cf. Figure 11.3(b)) and thereby reduce the number of elements. The boundary termed the bandwidth (see Figure 11.5). In an efficient FE program, advantage is
condition along the symmetry line is that the flux q. = 0. taken not only of the symmetry, but also of the banded structure of the K-matrix so
It is also often possible to make use of symmetry properties for more complicated that only the components of K within the bandwidth are considered when solving
situations. In Figure 11.4( a) we assume that heating cables are located at constant the system of equations Ka = f. Therefore, the bandwidth plays a significant role in
distances and that the isotropic body extends infinitely far away in the horizontal the computational cost of solving this system of equations and we shall now investigate
direction. Due to symmetry, the flux q. is zero along the planes AB and CD, which this aspect when the Gauss elimination technique is used.
226 Introduction to the finite element method Element meshes and global nodal numbering 227
bandwidth ( = 5) Elimination of a1 in the second system of equations yields
K = X X X 0 x' ,0 0 0 0 Kua1 +A1 a =f1; 1K1a= 1f ( 11.2)
X X X X X x', 0 0 0
X X X X 0 X 'x'-.. 0.... 0 where
0 X X X X X X x',~
,x X 0 X X X X 0 X
(11.3)
0'' X X X X X X X X
0 ()"",X X X X X X 0 This elimination process is called a static condensation. The derivations above require
0 0 0' ,X 0 X X X X that K 11 ¥ 0. However, the system of equations Ka = f refers here to that system
0 0 0 0 . . . ,x X 0 X X which is obtained after proper boundary conditions have been invoked. This implies
Figure 11.5 Illustration of bandwidth for symmetric matrix K that K is positive definite and, consequently, all diagonal terms are positive (cf.
page 24).
It appears that the two systems given by (11.2) may be expressed as

11.2.2 Systematic Gauss elimination ( 11.4)

The principal ingredients of the Gauss elimination technique were described in

Chapter 2. However, we shall now present a formulation of this technique, which where OT = [0 0 . . . 0]. It is evident that (11.4) consists of the first step in the
facilitates its implementation in a computer program. We recall that Gauss Gauss elimination process in which the unknown a 1 is eliminated from all the
elimination comprises a systematic elimination of unknowns so that a triangularized equations except the first.
form is achieved (cf. (2.56)- (2.58)). Considering now the system of equations 1K 1a = 1f we perform the same kind
Consider the equation system Ka = f and perform the partitioning shown below of elimination process a s indicated by ( 11.1 )- ( 11.4 ), i.e. the unknown a 2 is eliminated
I
... Kin from all the equations of 1 K 1a = 1 f except the first. This procedure is continued until
Kl l I K l2 K l3 a! jl
--- - ~- - ---- ---- ---- the triangularized form of the system of equations is obtained. It appears
K2! I K22 K23 K2n a2 fz that this approach to Gauss elimination in reality consists of a sequence of static
I
KJ I I K 32 K 33 K3n a3 condensations.
= /3 (11.1)
I To illustrate this approach, we consider the following system of equations:
I
I I
Knl I
Kn2 Kn3 Knn a. 2 I - 2 -2 0 a1
I fn - - - - 1----- ---- -
I
With the following definitions -2 I
4 -2 - 2 a2 0
I (11.5)
- 2 - 2 12 - 2 a3 - 5
K 21 K22 K23 K2n az !2
0 -2 - 2 22 a4 7
AT=
K 31 K 32 K33 K3n a3 /3
K= Ia = f=
With ( 11.3) we obtain
Kn1 K.2 Kn3
the system of equations ( 11.1) can be written as
Knn an f.
,K ~ -~[
-2
12 -2] [-2]
~~ -~ -~ [-2 -2 OJ~ [ -~
- 4
10
-2]
- 2
- ·2 - 2 - 2 -2 22

Carrying out the matrix multiplications we obtain

Kua 1 + A 1a=f1; ATa 1 +~ 1 a=f
,r~ HJ-~ f~]~[ -~] x
228 Introduction to the finite element method Element meshes and global nodal numbering 229
'The analogue to ( 11.4) then reads
that the establishment of
2 - 2 - 2 0 al K j 11 AT requires B - 1 divisions (11.9)
0 2 :- 4 -2 a2
0
---,- -- --
- 4 I 10 - 2 a3
--4-1 (11.6)
Referring to ( 11 .3) we need to establish the matrix K !/AT A. The matrix AT A consists
of(B- 1 )2 components, but as ATA is symmetric, we only need to compute tB(B- 1)
I components. That is, knowing the quantity K ! /AT, the determination of
0 - 2 ,- 2 22 a4 7
Using ( 11.3) gives K !/ AT A requires ! B(B - 1) multiplications (11.10)

2K =[
- 2
-2] 1[-4]
10
22
- -
2 - 2
[ -4 -2] = [ 2
- 6
-6]
20
Knowing this quantity, the determination of
K - K!/ ATA requires ! B(B - 1) subtractions (11.11)

= [-4] -21 [-4] = [-2]

Let us now define an operation as
2f 7 X 1 -2 8
one operation = one multiplication + one subtraction (11.12)
i.e. the following system of equations can now be established:
Moreover, let us disregard the computational effort related to the determination of
2 - 2 - 2 0 1 Ki/ AT and given by ( 11.9). It follows that the total computational effort for the
determination of
0 2 - 4 -2
0 0
0
(11.7) K- Kj/AAT requires !B(B- 1) operations. (11.13)
2 :-6 - 2
0 8 This effort relates to the elimination of one known. As we haven unknowns, we need
From ( 11.3) it follows that to make n - 1 eliminations, i.e. the total effort in the triangularization ofK becomes

total effort= (n- 1)tB(B- 1) operations (11.14)

3K = c2oJ - H -6J[ -6J = 2; 3r = [8J- tc -2)[ - 6J = 2

i.e. the final triangularized form becomes This expression is not completely true. We have assumed that the dimension of A is
1 x (B- 1 ), but in the later stages of the elimination process the dimension of A
2 - 2 -2 0 decreases, and at the last stage its dimension is only 1 x 1 ( cf. ( 11.7)). However, when
n » B we can ignore this aspect and accept the estimate given by ( 11.14 ). Moreover,
0 2 - 4 -2 1
if B » 1, (11.14) can approximately be written as
0 ( 11.8)
0 2 -6 - 2
0 0 0 2 2 I computational cost = anB 2 I (11.15)
in accordance with (2.57) and (2.58). Indeed, the reader is encouraged to compare where a is some parameter that depends on the computer. It is obvious that a small
the steps indicated by ( 11.5)- ( 11.8) with those arising from the elimination process bandwidth B enhances the computational efficiency. We emphasize that ( 11.15) is
described in Chapter 2. It appears that a complete similarity exists, but the advantage valid for the Gauss elimination · technique and that the computational cost for
of the method indicated here by ( 11.5)- ( 11.8) is its general and systematic formulation. other solution procedures cannot be estimated using ( 11.15). .
We mentioned previously that the size of the bandwidth of the system of equations In addition we mention that whereas the solution procedure presented above 1s
plays a significant role in the computational costs of solving the system Ka = f. Let based on a con~tant value of the bandwidth B - where Bin reality is the maximum
us now investigate this topic. As the triangularization of K is much more demanding bandwidth - other Gauss elimination techniques exist by. which the variation of
than the operations on the right-hand side f, we shall for convenience only consider
the bandwidth is considered . This variation forms a 'skyline' as indicated in Figure
the triangularization process of K.
11.6 and solution methods that take advantage of this aspect are called skyline or
Let the number of unknowns be denoted by nand the bandwidth by B. Then,
profile solvers. Such methods of solution as well as other approaches to the solution
referring to Figure 11.5 and (11.4), the dimension of A is 1 x (B- 1). We then have
of systems of equations are treated in depth by Bathe ( 1982).
230 Introduction to the h"nite element method Element meshes and global nodal numbering 231

8 = 5 6 3 2

K= X X i• 0 r---,
0 0 i .0... , 0 0
X I 0 x: 0 X ' 0
I I 0 Figure 11.8 Axially loaded bar modelled with simple linear elements
0 : 0 l0
·-- _J
X , x'I 0
' o: X ' 0
x•.... 0 obtain
I 3 4 5 6
--~
X X : o .Jr·x-·;o
L._ • I
skyline 2
X X 0 : 0 0 0 0
X X .. ___
:o K =1 k, - k, 0
2 - k, k, + k2 -k2 0 0 0
X X
X 3 0 - k2 k2 + k3 -k3 0 0
4 0 0 - k3 k3 + k4 -k4 0
Figure 11.6 Bandwidth B = 5 (in reality the maximum bandwidth). The variation of 5 0 0 0 -k4 k4 + ks - ks
bandwidth is indicated by the 'skyline' 6 0 0 0 0 -ks ks
(11.17)

11.2.3 Global nodal point numbering Instead of the numbering of the global nodal points as shown in F igure 11.7, we now
choose the numbering shown in Figure 11 .8. Using ( 11.16) again, we now obtain
The global stiffness matrix K and the global load vector f 1 are obtained as a summation
2 3 4 5 6
of all element stiffness matrices K• and all element load vectors ff, respectively, using
the topology data. This implies that the element numbers have no influence whatsoever K =1 k, 0 0 0 0 - k,
on the result or on computational efficiency. Likewise, as the topology data uniquely 2 0 ks 0 0 -ks 0
relates the local nodal numbers to the global nodal numbers, the local nodal numbers 3 0 0 k2 + k3 -k3 0 -k2
that we choose affect neither the result nor the computational efficiency. When we 4 0 0 -k3 k3 + k4 -k4 0
obtained the FE equations, we did not prescribe the global nodal numbers in any 5 0 - ks 0 - k4 k4 + ks 0
way. Therefore, the choice of global nodal numbers has no effect on the FE results 6 - k, 0 -k2 0 0 k, + k2
as such, but it does play a major role in computational efficiency because it influences (11.18)
the bandwidth B.
Let us illustrate this important fact by the axially loaded bar shown in Figure It appears that whereas the bandwidth of ( 11.17) is B = 2, the bandwidth of
11.7, where simple linear elements are used. Referring to (9.135), the element stiffness ( 11.18) is B = 6. In the present case, where n is a small number, we cannot use ( 11.15)
matrix K• can be written as directly, but it is obvious that the computational efficiency of the global nodal
numbering of figure 11.7 is much more efficient than that of Figure 11.8. We emphasize
Ke = k [ _ ~ -~] (11.16) that, corresponding to Figures 11.7 and 11.8, not only the global stiffness matrices,
but also the force vectors are changed in such a way that the two sets of FE results
become identical.
where the stiffness k is given by k = AE/ L and where it was assumed that AE is
constant along the element. The stiffnesses of the elements are also indicated in Figure
11.7. Following the procedures for establishing the global stiffness matrix K from the
element stiffness matrices, and using the global nodal numbering of Figure 11.7, we 11.3 Estimation of bandwidth
With this background, let us now identify a method of determining the bandwidth
B for an FE model. For one-dimensional heat flow, a component Kij of the global
2 3 I. 5 6 stiffness matrix K is given by (9.19), i.e.
k1 k2 k3 k4 ks bdN . dN .
Figure 11.7 Axially loaded bar modelled with simple linear elements
K- ·=
l)
f
a
- ' Ak - 1 dx
dx dx
( 11.19)
232 In tr oduction to th e finite el ement method Element meshes and global nodal n umbering 233
With reference to the discussion of this expression, we recall that K ii is zero unless
both nodal points i andj are present in at least one element. Considering all elements
in the body, let R denote the largest difference between the global nodal points i and
j found in an element; that is,

I R = Ii - j Imal( found in an element I (11.20)

where Ii - j I is the numerical value of the difference i - j and where R refers to the
largest difference found in all elements of the body. Considering the component K ;i,
it follows that if Ii - j I > R then the global nodal points i and j are never present in
the same element. Thus
Kij = 0 if Ii - j I > R ( 11.21)
Referring to Figure 11.5, we conclude that the bandwidth B is given by
B=R+1 (11.22)
This result was obtained for one-dimensional heat flow and it is also of interest to
study two-dimensional heat flow. For this purpose consider a component K ii of K
as given by (10.15). With (10.9) and (6.6) it is easily shown that

IJ
f
K..= [oN; (k oNi + k oNi) + oN; (k oNi + k oNi)] t dA (11.23)
A
!'>
uX
xx !'>
uX
xy !'>
uy
!'>
uy
yx !'>
uX
yy !'>
uy

Just as for (11.19), we draw the conclusion that Kii is zero unless both nodal points
i and j are present in at least one element; that is, we are again led to (11.21) and
thus to ( 11.22 ).
Indeed, this result is not surprising. We recall that a global shape function N;
differs from zero only in those elements which contain nodal point i. Therefore, if we
increase the temperature in nodal point i and leave all other nodal temperatures
unchanged, this increase only affects those elements that contain nodal point i.
T herefore, the temperatures in nodal points i andj are uncoupled unless i and j belong
to the same element. Consequently, K ii = 0 unless i and j belong to the same element. Figure 11.9 Different globa l nodal point numbering and effect on R
We conclude that ( 11.21) and thus ( 11.22) apply not only to one- and two-dimensional
heat flow, but also to three-dimensional heat flow.
Up until now we have studied heat problems where one degree of freedom, i.e.
one unknown, is coupled to each nodal point. We shall later encounter problems
where more unknowns are related to each nodal point and we immediately generalize ...... ......"' ......_:'">. ......_
.

......_......_,........_........__
( 11.22) to ...,[\
[\'\
......_ "' ...... ...... ......_

I B = (R + 1)NooF ( 11.24) ...,

. . , ~"-""
""
where N ooF is the number of unknowns (degrees of freedom) for each node.
In order to illustrate the applicability of ( 11.24 ), consider the two different nodal
point numberings shown in Figure 11.9. The largest difference R between two nodal
numbers in an element is found to be R = 9 in Figure 11.9(a) and R = 35 in 11.9(b).
\ Figure 11.10
""

One-, two- and three-dimensional models

"", [\
234 Introduction to the finite element method

Table ll.l
No. of equations, n Bandwidth, B Computational cost, cmB2
12
1-D 5 2 oc20 Stres es and st(ains
2-D
3-D
25
125
~7
~32
oc1225 .-:-!J {j:-':)._
oc128 000

Using ( 11.24) the respective bandwidths become 10 and 36, if there is one unknown
(NooF) at each node, and 20 and 72 if there are two unknowns at each node.
Finally, let us illustrate that an increase of dimension for a problem increases
the computational costs dramatically. Considering heat flow, Figure 11.10 shows
one-, two- and three-dimensional models with n = 5, n = 5 x 5 and n = 5 x 5 x 5
nodal points, respe~tively. The bandwidth B for these problems is given approximately In the previous chapters, the FE method was formulated for problems having one
by Table 11.1. Thts table clearly shows that apparently simple three-dimensional unknown function - the temperature. Apart from their technical importance, such
problems present a challenge even for large computers. problems present the most straightforward method of introducing the concepts of
FE analysis. We shall now shift our focus of interest towards solid mechanics problems
where, in general, we have more unknown functions, namely the displacements in
the x-, y- and z-directions. To obtain a firm grasp of the differential equations that
we will encounter, we will start with some introductory remarks on stresses and
strains. For a more comprehensive treatment of stresses and strains, we may refer to
Fung ( 1965 ), Mal vern ( 1969 ), Spencer ( 1980) and Timoshenko and Goodier ( 1970 ).

12.1 Stresses -- 1J
The body is assumed to be continuous and two kinds of forces are assumed: body forces
(i.e. force per unit volume) and surface forces (i.e. force per unit area).
Consider a surface of the body as shown in Figure 12.1. This surface can be an
external surface or an internal surface obtained from a section of the body. The vector
n is a unit vector normal to the surface and directed out of the body. The incremental
force vector dP acts on the infinitesimal surface area dA. When dA approaches zero,
it is assumed that the ratio dP I dA approaches a value given by

~he~ co:p: :~, t:.:,[;)t,in the and '~ir~tion<, respect:~:~ :

x , y-
is termed the traction vector and has the dimension [N/m 2 ]. The traction vector t
defined above is related to a surface with the outer unit normal vector n. It is obvious
that the traction vector will, in general, be different when other sections through the
same point are considered. What we look for is a quantity - the stress tens9r - which
f;:b ->~'-:(
235
236 Introduction to the finite element m ethod Stresses and strains 237
where Uzx• a,Y and u,z denote the components of Sy in the x-, y- and z-directions,
respectively (cf. Figure 12.2).
The components given by (12.2)- (12.4) are termed the stress components and
i;,
u xx• u >'Y' Uzz are called normal stresses, whereas uxy• uxz• Uyx • Uzx • Uz y are referred
to as sh_ear stresses. Weo bserve -the consistent notation of the stress components
Figure 12.1 Force dP on area dA with outer unit normal vector n wher(for instance, u yz is the z-component of the traction vector for a surface with
the outer unit vector in the y-direction, whereas u xx is the x-component of the traction
Ozz vector for a surface with the outer unit vector in the x-direction.

~a,,
Using the special traction vectors considered above, we define the matrix S as

( 12.5)

Figure 12.2 Illustration of stress components

This matrix is called the 'stress tensor. Here we need not probe into the concepts of
for a particular point contains all the information necessary to determine the traction
vector for arbitrary sections through the point. -
tensors, but simply mention that tensors are quantities that change in a particular
manner when the coordinate system is changed. The only topic of interest here is
•
Let us now consider some special traction vectors, namely those obtained when that S contains all the stress components.
sections perpendicular to the coordinate axes are considered. Assume that the outer Let us now prove that Sis symmetric. From the body we cut a small parallelepiped
nor~al vector n_(see Figure 12.1) is in the direction of the x-axis. The corresponding with planes parallel to the coordinate planes. We then consider the moment
tractwn vector IS denoted by sx and we can resolve this vector into its components equilibrium about an axis through the centre E of this parallelepiped and parallel to
along the coordinate axes, i.e. the z-axis ( cf. Figure 12.3 ). It appears that body forces do not provide a moment
about this axis. It is also obvious that only forces acting on planes parallel to the
moment axis can contribute to the moment equilibrium. On these planes, only shear
stresses normal to the moment axis can give rise to moments; see Figure 12.3.
, ( 12.2) Referring to this figure, the positive direction of the shear stresses along BC and
DC is in accordance with the previous interpretation of the stress components ( cf.
where ~xx• axy and axz denote the components of sx in the x-, y- and z-dirt>-::tions, Figure 12.2). The positive direction of the shear stresses along AB and AD follows
respectively. These components are illustrated in Figure 12.2. from the law of action and reaction. Taking moments as positive in the
Likewise, if the outer ~ormal unit vector n is taken in the direction of they-axis,
we denote the correspondmg traction vector by s>', i.e.

( 12.3)

where u yx • u JIY and uyz denote the components of s in the x- y- and z-directions y
• y ' '

.L,
respectively (cf. Figure 12.2). Finally, if the outer normal unit vector n is taken in
the direction of the z-axis, we denote the corresponding traction vector by sz, i.e.

( 12.4) I· dx

Figure J2.3 Moment a bout a n axis through the centre E and parallel to the z-axis
238 Introduction to the finite element method Stresses and strains 239

counter-clockwise direction, moment equilibrium about point E yields The components of the vector n are given by
(axy + daxy) dy dz!dx - (ayx + dayx) dx dz!dy
+ O"xy dy dz!dx- O"yx dx dz!dy = 0 ( 12.8)
i.e.
2axy - 2ayx + daxy- dayx =0 Denoting the angle between n and the x-axis by (n, x ) we find, for instance, that
cos(n, x) = nx. The area ABC is denoted by dA, the area AOC by dAx, the area AOB
Letting dx, dy and dz approach zero, both daxy and dayx also approach zero; that
by dAY and the area BOC by dAz. In Figure 12.4(b ), the line CP is orthogonal to
is, moment equilibrium requires that axy = ayx· Likewise, considering moment
the line AB. However, as n is perpendicular to the surface dA, it follows that n is
equilibrium about axes parallel to the x- and y-axes implies that ayz = azy and
located in the plane defined by OCP. Consequently
axz =am respectively. In conclusion, moment equilibrium requires that
dA y = ! OP.AB; OP = CP cos 8; dA = tCP .AB
( 12.6)
whereby
Referring to (12.5) we conclude that S is symmetric, i.e.
dAY= dA cos 8; cos 8 = ny; i.e. dAy = ny dA
(12.7)
With this and analogous arguments we find that
Our aim was to establish a quantity - the stress tensor - which contains all the dAx = nx dA; dAY= ny dA ; dA z = nz dA ( 12.9)
information necessary to determine the traction vector t for arbitrary sections through
The condition of force equilibrium of the tetrahedron of Figure 12.4(a) requires that
the point in question. We shall now prove that S contains this information.
Consider the infinitesimal tetrahedron shown in Figure 12.4(a). At the surface ( 12.10)
ABC with the outer unit normal vector n, we have the traction vector t. On the
planes parallel to the coordinate planes the traction vectors are -sx, - sY and -sz where the body force b is the force per unit volume and d V is the volume of the ,.
( cf. ( 12.2)- ( 12.4 ); minus signs appear because of the law of action and reaction and infinitesimal tetrahedron. The body force b has the components
because the outer normal vectors are in the negative direction of the coordinate axes).
I
(12.11 )
y

a) Using (12.9) in (12.10) gives

dV
t-sX nX -sy ny -sz nz + b -
dA= 0

Letting the size of the tetrahedron shrink towards zero, we have d V IdA -+ 0, i.e. we
obtain
( 12.12)

This may be written as

z
Figure 12.4
z
(a) Traction vectors on tetrahedron: t acts on ABC, - s" on AOC, - sY on AOB
and -sz on BOC ; (b) determination of dA y by geometrical arguments. Vector n is located in
the plane OCP
Hs. s, ·{}sT.
240 Introduction to the finite element method Stresses and strains 241

where (12.5) and (12.8) were used. AsS is symmetric (cf. (12.7)) we conclude that As this expression holds for arbitrary regions V, it is concluded that

I t = Sn ( 12.13)
div Sx + bx = 0
Inserting the expression for sx as given by (12.2) and using the definition of the
( 12.18)

This expression proves that knowledge of the stress tensor S provides sufficient divergence of a vector (cf. (5.30)) give
information for the traction vector t to be derived for any direction n. Equation
0(JXX 0(J XY 0(JXZ b
(12.13) is occasionally referred to as Cauchy's formula from 1822. It should be observed - - + - - + -- + x = 0
that on the exterior surface of the body, {12.13) represents a boundary condition ax oy oz
expressing a relation between the forces on the external surface and the stress tensor. Treating the two last equations of( 12.17) in the same manner, we obtain the following
With (12.1), (12.5) and (12.8), (12.13) can be written as differential equations :

0(J XX 0(Jxy 0(JXZ b

- - + - - + - - + x= 0
(12.14) ax oy az
Oayx OUyy Oayz b _ O (12.19)
-+ - +- + -
ox oy az y
Moreover, use of ( 12.2)- ( 12.4) yields the alternative formulation
oazx oazy oazz b 0
( 12.15) - - + - - + - + z=
ax oy oz
Instead of the infinitesimal tetrahedron of Figure 12.4, we now consider
equilibrium for an arbitrary part of the body. The forces acting on this arbitrary These differential equations express the equilibrium condition (i.e. the balance principle)
body are given by the traction vector t along the boundary surface S and the body for the body.
force b in the region V. Equilibrium requires that Let us define the following matrices:

1 t dS + t b d V = 0 (12.16)
a
ox
0 0
a a
oy az
0
(Jxx
(J yy
This expression comprises three equations which become
VT=
a a o a a=
(J zz
( 12.20)

t
0 0
oy ox oz (Jxy
f/x dS + bx dV = 0 a a a (J XZ
0 0 0
oz ax oy (Jyz
f/Y dS + f/YdV = 0 ( 12.17)
where a contains all the stress components and V is a matrix differential operator.

1 tz dS + t bz d V = 0
With these definitions ( 12.19) can be written in the compact matrix form

(12.21)
With tx given by ( 12.15), the first equation of ( 12.17) may be written as

1 L s!n dS + bx dV = 0
12.1.1 Plane stress
Plane stress is defined as a stress state where the only non-zero stresses are t1xx> a YY

The first term can be reformulated using Gauss' divergence theorem ( cf. (5.32)) to and aXY' i.e. the stress tensorS becomes

+~: :~: ~]
obtain

t(divsx + b,J dV =O
s (12.22)
242 Introduction to the finite element method Stresses and strains 243
The boundary conditions (12.14) then take the form located in the xy-plane. Thin disks loaded by forces in the plane are examples of
structures which may be analyzed using an assumption of plane stress (cf. Figure
12.5). In addition to the assumption above, it is also assumed that nothing depends
• (12.23) on the z-coordinate, i.e. plane stress constitutes a two-dimensional problem.

and
t. = 0
12.2 Strains -y
( 12.24)
/ I J' "'
It is important that plane stress conditions require that the component t. = 0. Having_ol)Jained ..a description of the stresses in the body, we next consider its
Moreover, we may write @efor~atio~ This deformation manifests itself as a change of distance between two
rieiglrbouring material points and as a change of angle between two intersecting lines.
t = [::J [::J
n = ( 12.25) We will now proceed to quantify these phenomena.
Before deformation, a point in the body is described by the coordinates (x, y, z).
Finally, the equilibrium conditions (12.19) become After deformation, this point has moved so that it now has the coordinates (x + u,,
y + uy, z + u.). The changes ux, uY, u. caused by the deformation are termed
displacement components and we collect these components in the displacement vector
u given by
( 12.26)

(12.29)

and it is required that b. = 0. Defining the matrices

The displacements of a point (x, y, z) are given by u = u(x, y, z). For a neighbouring
point (x + dx, y + d y, z + dz), the displacements become u + du. Using the chain
rule, du is written as
( 12.27)
OUx OUx OUx
dux = - dx +- dy +- dz
ax ay az
( 12.26) can be written as OUy OUy OUy d
du = - dx +- dy +- z ( 12.30)
I \?Tu + b = 0 ( 12.28)
y ax ay az

OUz OUz au.

It was observed that plane stress requires that t. = 0 and b, = 0, i.e. both the du =- dx +- dy + - dz
• ax ay az
traction vector t along the boundary of the body and the body force b must be forces
A derivative like ouxf ox is called a displacement gradient.
llefore any deformations, consider next the line AB parallel to the x-axis (see
Figure 12.6). After deformation this line takes the position A'B'. Referring to this
figure, the length IAB I of AB is
IABI = dx (12.3 1)
Likewise the length IA'B' I of A'B' becomes
IA'B' I = [(x + dx + Ux + dux- X - ux) 2 + (y + Uy + duy - y - uy) 2
Figure 12.5 Thin disk loaded in plane stress. Traction vector t and bc·dy force vector b are
located in the xy-plane + (z + u. + du. - z- u. )2 ] 112
244 Introduction to the finite element m ethod Stresses and strains 245

s'
A~x+u,+dux. y+Uy+dUy .Z+U 2 +du 2 )
(x+Ux,y+uy . Z+Uz)
y

A B
(x ,y,z) (X+ dx.y.z) y

/------x

dx
z
Figure 12.6 Deformation of line AB into line A'B' '-----x
Figure 12.7 Distortion of the right angle CAB due to deformation
i.e.
( 12.32)
As the line AB is parallel to the x-axis, we have dy = dz = 0, i.e. ( 12.30) gives
This relation was derived for a line parallel to the x-axis and in accordance with the
oux OUy au. usual terminology we denote this relative elongation as the normal strain exx in the
du = -dx· du = -dx· du = - dx
X OX y ox ' z ox x-direction, i.e. exx = oux/ ox. Treating lines parallel to the y- and z-axes in the same
manner, the following normal strains are derived:
Use of these expressions in (12.32) yields

IA'B'I = dx[(1 + ~~ r (~;r (~~YJ'

+ +
2

This expression is exact. In most engineering applications, however, the displacement

( 12.33)
_, i ;(,
)j:-fJ-~
( 12.37)

Next let us evaluate changes of angles. Before the deformation, two orthogonal
gradients are small when compared with unity, and, as examples, we have
lines AB and AC parallel to the coordinate axes are considered (see Figure 12.7).

OX I« 1; IOUy
ox I« t; Iou,
OX I« 1
Due to the deformation, points A, B and C move to A', B' and C', respectively, as
Ioux ( 12.34) shown in the figure.
As we are only interested in changes of angles~ and as small displacement gradients
The assumption of small displacement gradients implies that are assumed, we can ignore changes in the length of AB and AC, i.e. IA'B' I = IAB I
and IA'C' I = IACI. Therefore, Figure 12.7 provides that

. du du . dux dux
sm ~1 = I A'~' I (12.38)
which simplifies ( 12.33) to = d;; sm 82 = IA'C' I = dY
IA'B'I = dx( 1 + ~~) ( 12.35) Along AB, dy = dz = 0 holds, i.e. ( 12.30) yields duy = ( ouy/ ox) dx. Likewise, along
AC, dx = dz = 0 holds, which implies that dux= (ou xfoy ) dy. Use of these expressions
in (12.38) gives
where 1 + auxfox is positive (cf. (12.34)). We are now in a position to calculate the
relative elongation of the infinitesimal line AB. From ( 12.31) and ( 12.35) it follows that e- OUY.
1
( 12.39)
IA'B'I-IABI oux - ox'
( 12.36)
IABI ox where sin e ~ () for small angles. It follows that the orthogonal angle CAB has
246 Introduction to the fin ite element method Stresses and strains 247

decreased due to the deformation by the amount

ouy oux
et+ e2 =
)-.
-+-
ox oy
This amount is called the shear strain Yxy> where the subscripts indicate that lines
AB and AC are parallel to the x - andy-axes, respectively. Treating changes of angles
between the other coordinate directions in the same manner, the following shear z
strains are derived :
Long body loaded in plane strain. Traction vector t and body force vector b are
tr>-0._ (12.40)
Figure 12.8
located in the xy-plane and do not depend on the z-coordinate

equilibrium condition ( 12.21) and in { 12.42). As {12.42) is based on purely kinematical

It appears that the shear strains are symmetric; for instance, we have Y x y = Yyx· considerations, it is also referred to as the kinematic relation.
With the normal strains { 12.37) and the shear strains {12.40), a complete
description of the deformation of the body has been achieved. As the strains are
obtained as derivatives of the displacements, the strains are independent of rigid-body 12.2.1 Plane strain
motions - as we would expect. However, we emphasize that the strains above were
derived under the assumption of small displacement gradients, i.e. small strains. In A state of plane strain exists if the only non-zero strains are exx> eyy and Y xy and if
practice, small strains mean strains smaller than, say, 3- 5%. For large displacement nothing depends on the z-coordinate. This is equivalent to the following displacements:
gradients, proper strain measures become much more involved than {12.37) and ux=ux(x,y); Uy =uy(x,y) ; u,=O (12.43)
(12.40) and we may refer to Fung (1965), Malvern (1969) and Spencer (1980) for a
treatment of these matters. As nothing depends on the z-coordinate, plane strain constitutes a two-dimensional
Let us define the following matrices : problem. Let us define the matrices

a a 0
0 0 ox
OX

e xx

Eyy
0
a
ay
0

a
ll =
l'"]
Eyy

Yxy
;
u= [::l ~= 0

a
0
oy
0
{12.44)

e,,
0 0
oz oy ax
t= V= ( 12.41)
'Yxy a 0
0
Then the kinematic relations ( 12.37) and ( 12.40) can again be written as
oy
Yxz

a
ox
0
I t = ~u I ( 12.45)
'Yyz
0
oz ox We note that the matrix differential operator V defined above is the same as that
entering the equilibrium differential equations (12.28) for plane stress.
0 Plane strain often occurs in practice when a long prismatic or cylindrical body
0
oz oy is loaded by forces which are perpendicular to the longitudinal axis and which do
where e contains all the strains and ~ is a matrix differential operator. With these not vary along this axis. In this case it can be assumed that all cross-sections are in
definitions ( 12.37) and ( 12.40) can be combined into the compact form the same state and if, moreover, the body is restricted from moving in the length
direction, a state of plane strain exists. An example is a long retaining wall with
= ~u
lateral pressure (see Figure 12.8). We observe that plane strain and plane stress may
I t I (12.42)
be considered as two extreme and opposite conditions: whereas the first is applicable
It is of considerable interest that the matrix differential operator ~ enters both the for long bodies, the latter may hold for thin bodies.
 Linear elasticity 249
l3 ____________~ of linearity between stresses and strains, is given by

r Linear elasticity J I
where
u =De

(j XX
I
llxx
( 13.2)

Ciyy Dll D12 D16 llyy

(j zz D21 D21 D26 llzz

D= ll= ( 13.3)
t1=
(i xy Yxy

(j xz D61 D 62 D 66 Yxz
(j yz Yyz
In the previous chapter, we established the concepts of stresses and strains. No
and Dis the constitutive matrix. As linearity is required, the matrix D is constant for
reference was made to the material as such, and we emphasize that within the
a given position. Occasionally, (13.2) is referred to as Hooke's generalized law. This
ass~mption of small strains the results of Chapter 12 hold for any material
which may be treated as a continuum. It is obvious, however, that stresses and strains generalization is similar to that of Fourier's law from one to three dimensions ( cf. ( 4.3)
must be related in some way or another and the specific manner of this relation is and ( 6.4)). The elasticity defined by ( 13.2 ), which possesses the properties of linearity
controlled by the specific material in question. The relation between stresses and and a one-to-one relation between stresses and strains, is also termed Cauchy elasticity
strains is called the constitutive relation and a variety of such relations has been (Cauchy, 1789 - 1857). If, in addition to these properties, we also require that the
established. Examples are elasticity, plasticity, viscoelasticity, viscoplasticity and creep. strain energy for a given strain state only depends on the strain state itself and not
Here, we shall consider the simplest constitutive theory, namely linear elasticity. the manner in which this strain state was obtained, we obtain so-called hyperelasticity
In one dimension, linear elasticity is expressed by H ooke's law from 1676 or Green elasticity from 1839. In this case, the constitutive matrix Dis symmetric, i.e.

cr = Ee ( 13.1) ( 13.4)
where the material constant E is Young's modulus. This expression is illustrated in
Fi~ure 13.1, and we note that loading as well as unloading follow the same path.
as shown, for instance, by Malvern ( 1969) and Sokolnikoff ( 1946). The symmetry
This means that the material response is path independent; alternatively one says that property ( 13.4) is used almost universally within linear elasticity, and it will also be
the material response is history independent, as there exists a one-to-one relation adopted here. Due to this symmetry the 36 elasticity coefficients of D reduce to 21
between stress and strain. We also emphasize the linear response of the material. independent coefficients.
These characteristics also hold for linear elasticity with several stress and strain Let us investigate the concept of strain energy W per unit volume of the body
components. In this general case the stresses are given by u and the strains by e (cf. (see Figure 13.2), i.e. W has dimension [Nm/ m 3 ]. For a uniaxial stress state, the
( 12.20) and ( 12.41 )), and a direct generalization of( 13.1 ), which maintains the concept incremental strain energy is defined by

a dW =cr de; i.e. W(e)= tcr(e)de (13.5)

y---lood in g a a

dW=OdE

Figure 13.1 Linear elasticity Figure 13.2 Incremental strain energy dW and strain energy W

248
250 Introduction to the finite element method Linear elastici ty 251

Using Hooke's law ( 13.1) we obtain the familiar result the reader with some general information on different D-matrices that may be
encountered in practice and not to present a detailed discussion.
w = ~&2 ( 13.6)
Adopting this approach in the general case gives
dW =aT ds; dW = dsTa (13.7)
where dW = dWT. From Hooke's generalized law (13.2) it follows that
13.1 Symmetry properties
dW = sTD ds; dW= dsTDe ( 13.8)
It turns out that, in general, the D-matrix changes if another coordinate system is
where the symmetry property (13.4) was used. We may rewrite (13.8) in the form
chosen. A symmetry plane is said to exist if two coordinate systems, which are mirror
dW = ±( eTD de + dsTDs) = ~ d (eTDe) ( 13.9) images of each other with respect to this plane, leave the D-matrix unchanged. If no
symmetry planes exist the material is anisotropic (see Figure 13.3) and the symmetric
This gives
D-matrix contains 21 independent coefficients; see (13.3) and (13.4). If one symmetry

W(s) = J:dw = t~ d(sTDs)

plane exists and if this plane is parallel to the xy-plane (cf. Figure 13.3) it can be
shown that 13 independent coefficients exist as shown below:

where the limits of integration indicates that the integration should be taken from Dll D12 D13 Dt4 0 0
zero to the current strain state. This integration yields
D 21 D22 D23 Dz4 0 0
W = tsTDs I ( 13.10) D31 D32 D3 3 D34 0 0
(13.14)
D=
D4 t D42 D43 D44 0 0
The strain energy W for a uniaxial state of stress given by ( 13.6) shows that this
strain energy is positive (as we certainly expect Young's modulus E to be positive). 0 0 0 0 Dss Ds6
It seems natural also to expect that the strain energy is positive for the general case, 0 0 0 0 D6s D66
i.e. ( 13.10) gives
If three planes of symmetry exist and if the coordinate planes are parallel to these
(13.11) /

for all s # 0, which shows that D is positive definite. We may refer to the similar

Jif1l
result for heat flow (cf. (6.8)). As D is positive definite, we conclude from (2.67) that
D is non-singular, i.e.

I det D # 0 (13.12) .
a nisotrop y one symmetry p lane or t hot ropy
This means that (13.2) can be inverted to yield (th ree sy mmet ry pl ones )

I s= Ca; C = D- 1 (13.13)

where C is termed the compliance or flexibility matrix. After having discussed these
general properties of the constitutive matrix D, we shall now evaluate its components,
i.e. the elasticity coefficients. We will review certain material types and refer to
Lekhnitskii (1981 ), Love ( 1944), Malvern ( 1969) and Sokolnikoff( 1946) for a detailed
X
~I
transverse isotropy isot r opy
discussion. It is emphasized that the intention of this review is confined to provide Figure 13.3 Illustration of increasing degree of symmetry
252 Introduction to the finite element method Linear elasticity 253
planes, the D-matrix turns out to be applications. It is also recalled that the shear modulus G is given by

Du Dl2 Dl3 0 0 0
~ ( 13.18 )
~
D21 D22 D23 0 0 0
D31 D32 D33 0 0 0
D= (13.15) and with ( 13.2), ( 13.17) and ( 13.18) it follows that
0 0 0 D44 0 0
0 0 0 0 Dss 0 (13. 19 )
0 0 0 0 0 D66 From (13.2) and (13.17), we observe that the shear stresses and shear strains a re
independent of (i.e. uncoupled from) the normal stresses and normal strains and vice
i.e. we have nine independent coefficients. Such a material is said to be orthotropic versa. Previously, we required D to be positive definite ( cf. ( 13.11 )) and for an isotropic
and it has important practical applications, for instance for wood. It can be shown material it can be shown (e.g. Malvern, 1969, p. 501 ) that this is fulfilled if
that two symmetry planes imply the existence of three symmetry planes ( cf. Figure
13.3 ). I sotropy exists if the D-matrix is the same for all coordinate systems. If a plane I E>O ; -l<v<t (13.20)
of symmetry exists and if isotropy exists in this plane, one speaks of a transversely
isotropic material. If the plane is taken as the xy-plane, the D-matrix is given by It is straightforward to invert D for isotropic materials to obtain

Du 1 -v -v 0 0 0
D12 Dl3 0 0 0
D2 1 D22 0
-v - v 0 0 0
Dl3 0 0
0 1 -v -v 1 0 0 0
D=
D31 D31 D33 0 0 o - 1 =C= - ( 13.21)
0 0 0 !(Du - D12) 0 0
( 13.16) E 0 0 0 2(1 + v) 0 0

0 0 0 0 0 0 0 2( 1 + v) 0
0 Dss 0
0 0 0 0 0 0 0 0 2(1 + v)
0 0 Dss
We finally remark tha t if D depends on position, i.e. D = D (x, y, z), then the material
i.e. five independent coefficients. This type of material is illustrated in Figure 13.3 is referred to as inhomogeneous; otherwise it is termed homogeneous.
and is typical of stratified materials. Finally, if every plane is a symmetry plane, i.e. •
if the D-matrix takes the same form irrespective of the coordinate system, we have
an isotropic material. Here

1- v v v 0 0 0 13.2 Initial strains

v 1- v v 0 0 0
E v v 1- v From ( 13.13 ) it appears that zero stresses imply zero strains. However, there exist
0 0 0
D= situations with the presence of non-zero strains for zero stresses, i.e.
(1 + v)O- 2v) 0 0 0 !0-2v) 0 0
0 0 0 0
s = Ca + s 0 ( 13.22)
to - 2v) 0
0 0 0 0 0 t o - 2v)
where s 0 , the so-called initial strains, are the strains fo r zero stresses. As D = c- t,
(13.22) implies that
(13.17)
a = D (s - s0 ) ( 13.23)
i.e. two independent coefficients. These coefficients are Young's modulus E and Poisson's An important example of such initial strains is thermal strains, i.e. strains caused by
ratio v. We recall that an isotropic material has the same properties in all directions the thermal expansion of the material. Consider a specimen of isotropic material
(cf. Figure 13.3 ). This type of material is the one most often used in engineering which is free to expand as a result of a change in temperature. As the material is free
254 Introduction to the h"nite element method
.. Linear elasticity 255

to expand, t1 =
0 holds, and the thermal strains are then given by 'Yxz = 'Yyz = 0 ( 13.29)

Let us define the following matrices:

e0 = iXLlT
0
(13.24)
e=
[
6 ]
6xx '
yy ,
C = _!_
E
[
-
1 -v
V 1 ~ ]; t1 = r:::]; IXLlT[~]
e0 = (13.30)

'Yxy 0 0 2(1 + v) Uxy 0

0
Then ( 13.27) can be written in the compact form
0
(13.31)
e= + e0
where IX is the thermal expansion coefficient [ 1;oc] and .::l T is the change in
Ct1

temperature. In accordance with the assumption of isotropy, the thermal normal which leads to
strains are equal and no thermal shear strains exist. With ( 13.17) and ( 13.24), ( 13.23)
can be written as I a = De - De 0
( 13.32)

I t1 = De - De 0 ( 13.25) where

where
1
D=C-l=_E_r: 1 - v2
v ~ ] (13.33)

0 0 ! (1 - v)
1
and
Deo = 1XE.::lT 1
( 13.26)
1- 2v 0
0
iXELlT
De0 = - - 1
1- v [1] (13.34)

0 0
When thermal strains are considered in Hooke's law ( 13.25) one speaks of It is emphasized that the in-plane StresseS U xx• UYY' U xy directly determine the in-p/ane
thermoelasticity. strains 6xx' 6YY' '\)1 xy and vice . versa. Moreover, when the in-plane stresses are known,
these stresses determine the remaining strain components (cf. ( 13.28) and ( 13.29 )). We
note in particular that the out-of-plane strain 6zz• in general, is different from zero.
13.3 Plane stress
For plane stress the only non-zero stresses are uxx, uyy and uxy (cf. (12.22)). Using
t~ese conditions in ( 13.22) and assuming isotropy and that e0 are the thermal strains
giVen by ( 13.24 ), we obtain from the first, second and fourth equations 13.4 Plane strain

[ :: ~ ~ -~
6xx] [ 1 -V

; 2(1
0 ][Uxx] + [1]
0
+ v)
Uyy

Uxy
iXLlT 1
0
( 13.27)
For plane strain the only non-zero strains are 6w 6yy and 'Yxy (cf. (12.43)). Using these
conditions in ( 13.25) and assuming isotropy we obtain from the first, second and
fourth equations
v
and from the third, fifth and sixth equations
1- v
(13.28) 0
 256 Introduction to the finite element method Linear elastiCity 257
•
and from the third, fifth and sixth equations The stresses a must fulfil the differential equations of equilibrium ( 12.21 ), i.e.
) Ev a.EAT
u•• = (e + e ) - - - ( 13.36) ( 13.43 )
•• (l+v)(1-2v) xx " 1 - 2v

(13.37) where b is the body force vector. The strains e must fulfil the kinematic relation
( 12.42 ), i.e.
Let us define the following matrices :

•
(Jxx]
~ [ ::: ; D ~ ( l h
[1-
)~H') ~
V
v
1- v 0 ].
where u is the displacement vector. Within our assumption of small strains, these
(1 3.44)

0 to- 2v) , equations hold for arbitrary solid materials which can be considered as continua.
(13.38) For a specific kind of material, the relation between stresses and strains is given in

•~ [;::J ~ ~ h)aA{] (!
terms of the constitutive relation. Here, we assume the constitutive relation in the
form of linear elasticity; that is, Hooke's generalized Jaw ( 13.25) states that

Then ( 13.35) can be written in the form

I a ,;, De - De0 ( 13.45)

I a = De- De 0 (13.39)
which is formulated here so that it includes the effect of thermal strains. The expressions
( 13.43 )-( 13.45) are generally referred to as the field equations for thermoelasticity.
These field equations have to be solved given certain boundary conditions. T hese
This relation leads to boundary conditions may be expressed in terms of prescribed traction vectors t or
e = Ca + e0 ( 13.40) prescribed displacements u. That is, for three-dimensional problems,..we obtain with
( 12.13)
where

t = Sn = h on Sh
-1 1+v[1 - v ( 13.46)
C=D =~ ~v ( 13.41) u=g on Sg

where h and g are given vectors. Sh is that part of the boundary S o n which the
With ( 13.40) it is possible to rewrite ( 13.36) as
traction vector t is prescribed and Su is that part of the boundary S on which the
Uzz = v(uxx + Uyy)- a.EAT (13.42) displacement vector u is given. The sum of Shand Su constitutes the entire boundary
S. We note that it is not possible to prescribe the traction vector and the displacement
As in the case of plane stress, we emphasize that the in-plane strains exx, eYY' Yxy
vector at the same position.
determine directly the in-plane stresses uxx• uYY' uxy and vice versa. When the in-plane
With these observations, the equations for solid mechanics may be summarized
strains are known these strains determine the remaining stress components ( cf. ( 13.36)
as shown in Figure 13.4.
and ( 13.37)). We note that the out-of-plane stress um in general, is different from zero.
In order to solve a specific problem, the number of unknowns must be the same
as the number of equations. Let us check whether (13.43)- (13.45) fulfil this
requirement, noting that the body forces band the thermal strains e0 are prescribed.
For three dimensions (3-D), we recall that ( 13.43) comprises three equations, ( 13.44)
six equations and ( 13.45) six equations. A survey of equations and unknowns is given
13.5 Summary of equations for solid mechanics in Table 13.1.
It is no surprise that exact analytical solutions of three-dimensional elasticity
It may be advantageous to summarize the relations established so far for problems present a formidable task, and as a consequence they are scarce. Moreover,
three-dimensional solid mechanics. these solutions are confined to physically simple problems with uncomplicated
258 Introduction to the finite element method Linear elasticity 259
•
stresses 13.5.1 Comparison of the equations of heat flow and solid
mechanics

Having established the equations governing solid mechanics, it may be of interest to

constitutive make a comparison with the equations of heat flow. The result is given in Table 13.2.
relations:
elasticity oun ary The variable T or u is often termed the 'state' variable as it comprises the
(u=Dt)
pl asticity
creep
+ conditions
t = h on sh
u = g on S
unknown, i.e. the state ' itself. The distribution of the state variable gives rise to a
certain physical quantity called the 'flux' vector (q in heat flow, a in solid mechanics).
At the boundary, this flux vector is determined by q. for heat flow and by the traction
vector t in solid mechanics. It is characteristic that the divergence always enters the
balance principle, which takes the form of a conservation principle in heat flow and
an equilibrium condition for solid mechanics. In order to relate the state variables
Tor u to the fluxes q or a, a 'kinematic' quantity is derived from the state variable.
In heat flow, this ' kinematic' quantity is the temperature gradient VT whereas in
solid mechanics it is the strain e. The flux and kinematic quantity are related by the
Figure 13.4 Illustration of fundamental equations of solid mechanics constitutive law in terms of Fourier's or Hooke's law. Finally, the similarity of the
boundary conditions is also apparent from the table.
We note that insertion of the constitutive law into the balance equation gives_
Table 13.1 Summary of equations and unknowns
Number of equations div(DVT) + Q= 0 ( 13.47)
Type of equation 3-D 2-D 1-D
for heat flow, whereas for elasticity we obtain
Equilibrium VTa + b = 0 3 2
Constitutive relations u =De- De0 6 3
Kinematics e = Vu 6 3 ( 13.48)
Total 15 8 3

Number of unknowns
Type of unknown 3-D 2-D 1-D Table 13.2 Equations of heat flow and of solid mechanics

Stresses (1 6 3 Heat flow

Strains e 6 3 (scalar field Solid mechanics
Displacements u 3 2 problem) (vector field problem) Quantity Remarks

Total 15 8 3 T u state variable

q (1 flux vector
q" = qTn t = Sn flux at boundary
div q = Q divsx + bx = 0}
geometry. The advantage of the FE method is that we can in principle solve any
d!v sy + by : 0 VTa +b= 0 balance or conservation
problem, irrespective of the geometry and even considering complex constitutive d!V Sz + bz - 0 principle
relations. Here we confine ourselves to thermoelastic problems, but much more' VT e = Vu 'kinematic' quantity
involved constitutive models like, for instance, plasticity and creep can also be treated q = - DVT u =De constitutive law -
by the FE method (cf. for instance Zienkiewicz, 1977). The expense that we pay for Fourier and Hooke
this gain in generality is that the FE method provides an approximate solution, T = g on S 9 u = g on S9
which, however, converges towards the exact solution for a decreasing element size. qn = h on sh t = h on sh boundary conditions
260 Introduction to the finite element method

This last set of equations is termed Navier's equations from 1821. For orthotropic and
14 _ _ _ _ _----.
isotropic materials, all the terms appearing in (13.47) are given by (6.38) and (6.39).
In a similar manner, the reader is invited to identify all the terms in Navier's equations
( 13.48) and it will turn out that, even for isotropic elasticity, the result is highly
FE formulation of torsion of
complicated. T his supports the fact already stated that exact solutions of the field
equations of elasticity are scarce. non-circular shafts

We have emphasized the difficulties of obtaining exact solutions to the field equations
of elasticity. Now we shall consider one of the exceptions for which we may obtain
an advantageous reformulation of these equations that allows for exact solutions for
some geometries and that provides the basis for an FE analysis for general geometries.
~!'fl.. tfhe problem we consider is that of torsion of non-circular elastic shafts, i.e.
~ prismatic members. Whereas the solution of circular shafts is trivial, this is certainly
not ilie.case[~; 'Mn-circular shafts. The response of such shafts is of great technical
importance, and the theory presented below is essentially indebted to Saint-Venant
in 1855. This so-called Saint- Venant torsion theory is also of historical significance,
since it provides the first exact solution to the field equations of elasticity theory.
Moreover, it is of consider able interest that Saint-Venant torsion theory leads to
equations that are similar to those of heat flow, and therefore we may take advantage
of the results obtained previously.

14.1 Saint-Venant torsion theory

14.1.1 Differential equation for Saint-Venant torsion theory __.\
. r'-7~\!J i:.:4tJ
The non-circular shaft shown in Figure 14.1 i~ suojected to a twisting mt ment (i.e.
torque T) about the axis~._.of1 .tpe prismatic m~mbeE: The cros'S:section is constant
along the axis and the torqtl'e~ is applied at the ends of the shaft. A coordinate system
~
is chosen so that the z-axis is in the axial direction of the shaft. The essential
characteristic of torsion of non-circular shafts is that they warp, i.e.~
displacements in the z-dir:_~ctiqn.Jhe theory below follows partly that g1ven, for
~ng(19()5)and Timoshenko and Goodier (1970). The key point is that
certain assumptions for the displacements are made and it is then checked that all
field equations, in fact, are satisfied exactly. ~ r-
A cross-section of the shaft perpendicuJaft~he z-axis is shown in Figure 14.2.
Apart from the z-axis being in the axial direction, the location of the coordinate

261
'
262 Introduction to the finite element method ' FE for mulation of torsion 263
T With ux and uY being the displacements in the x- and y-directions, respectively, we
0 then obtain

I ux = x'- x = -q;(y - y 0 ); Uy = y' - Y = q;(x- x 0 ) (14.4) )\

U;-
Since it was assumed that the projection of the cross-section in Figure 14.2 rotates
as a rigid body, the rotation angle q; can only depend on z, i.e.
v~
Figure 14.1 Torsion of non-circular shaft (;1,1

y
q; = q;(z)

Having determined the displacements ux and u Y, we shall now iqy-e"S~ate the

(14.5)
·f J
r
displacement Uz in the z-direction. The displacement uz is called the [warping and it
is characteristic for non-circular shafts that they warp, i.e. uz # 0. We ~onsider a shaft
for which the cross-section does not change with position along the axis and the
shaft is loaded by opposite torques at the ends of the shaft. O n this background, it
seems reasonable to assume that the warping is independent of the z-coordinate, i.e.
Yo
Uz = u.(x, y) I (14.6)

z ~------------~----------------------x
We emphasize that the displacements given by (14.4)- (14.6) are assumed quantities.
However, it will appear that under certain circumstances, these displacements are, in
fact, the exact ones. To see this, we shall in the following prove that all field equations 1
J
Figure 14.2 Cross-section of member and centre of twist (CT) and boundary conditions are fulfilled exactly.
· J~ ~~[i)Iu)
system IS aru1tr'ary. t IS assumed t hat th e shape o f th.1s cross-sectiOn · ·
remams
From the displacements given by ( 14.4) and ( 14.6 ), the corresponding strains are '
obtained from the kinematic relations ( 12.37) and ( 12.40), i.e.
piicn~Ji.geajduring torsion; that is, the rn~ection of the cross-section is assumed to
rotate in tne !Y;J?.!.a.!!e as a ~~~od:{~bbut a centre, the so-called centre of twist
( 14.7)
(CT ), with the position (x 0 , y 0 ) ( cf. Figure 14.2). Before any deformation, we consider
the point P shown in Figure·14.2 with the coordinates Li•':P
. ~ tS· , :" u and
x= x 0 + rcosa; y=y0 +rsma ) l·
0
,_,.-l--f'J (14.1)
I
wh.eJ~ ris the distance betl,\'een P and..CI.~nd a is the angle shown in the figure. ( 14.8)
Due to the deforma tion," the p"'OintP rotates an angle q; about CT to the position P'.
As the projection of the cross-section in Figure 14.2 is assumed to rotate as a rigid
body, the distance from P' to CT is r. The coordinates of P' are Having determined the strains, we next consider the constitutive model which is
that of linear isotropic elasticity. From ( 13.2 ), ( 13.1 7) and ( 13.18) it follows that
x' = x 0 + r cos( a + q;); y' =Yo + r sin( a+ q;) ( 14.2)
AL!!sual, we assum~ small strains; that is, the rotation angle q; is small and thus ( 14.9)
Ieos q; ::::::: 1'and sin q; ::::::: q;. Using standard trigonometric relations, ( 14.2) then becomes

~0 + r_' f s aj - =~~~sin aj:~q; cos a

and
x' = rq; sin a; y'
which, due to (14.1) and Figure 14.2, can be written as (14.10)
~
x' = x- q;(y - y 0 ); y' = y + q;(x - x0 ) ( 14.3)

)\
} -
Yc ~~S·'C ~ ':.fvj 1 y g ~osc}.
I

- ,~
r ''jJ)
,. j ' 'S (f' 7")
 J
264 Introduction to the finite element method FE formulation of torsion 265
Finally, the last set of field equations to be invoked is the differential equations This is the quasi-harmonic differential equation that we have previously seen govern
of equilibrium given by (12.19). With (14.9), and as we have no body forces, these two-dimensional heat flow (cf. (6.26)). If G is constant, (14.17 ) reduces to Poisson ' s
/ differential equations take the form differential equation

0 (1xz = O· 0(1yz = 0 o2¢ o2¢

(14.11) -+ - + 2G8 =0 (14.18)
oz ' az ox 2 oy 2
and ( cf. ( 6.27)). With the definition
0(1xz
- + - =0
ox
0(1yz

oy
( 14.12) D = [1 / G 0 = 1
0 .1 / G G
J !._ (14.19)
'
\rG~G(x.y) l
We as~_ume that th<? shear modulus G does not depend on z, i.e. ( 14J 7) can be rewritten according to

"-- - /
(14.13) I div (DV¢) + 28 = 0 y ~Jl (14.20)

Use of(14,1U) in (14.11) then leads to where V¢ is the gradient d f ¢ (cf. (5.2)). We may also compare (14.20) with the
/ differential equation (6.32) fo r two-dimensional heat flow.
d2cp
-= 0 (14.14) At this point we have shown that a solution to (14.17) is in accordance with the
dz 2 exact fulfilment of all kinematic, constitutive and equilibrium equations.
since u. is independent of z (cf. (14.6)). From {14.14) we conclude that

dcp 14.1.2 Boundary conditions

8 =- = constant (14.15)
dz
To solve ( 14J 7) we need to establish proper boundary conditibns for the stress
i.e. the rate of twist 8 is constan~11~
the member. The dimension for 8 is [rad/ Il\J.
Let us,lOliowmg- Prandtl in 1903, define a so-called stress function ¢(x, y) by
function ¢( x, y ). As ¢ through ( 14.16) determines the stresses, it is no surprise that
' the boundary conditions for ¢ emerge from the stress conditions along the boundary
(/i'¥ the cross-section; see Figure 14.3. - - -
o¢ a¢
tp 't)-; )J- Let n denote the unit vector normal to the boundary and directed outwards from
qxz = -; qyz = -- ( 14.16) the body. Moreover, the unit vector m is tangential to the boundary and its direction
oy ox is chosen so that n and m, after a suitable rigid-body rotation, have the same directions

This definition has the important consequence that all the equilibrium differential
equations (14.11) and (14.12) are identically satisfied. Moreover, it turns out that the y i:jJ
problem of torsion can be expressed entirely in terms of the stress function ¢(x, y).
To see this, insert (14.16) in (14.10) and use (14.15) to obtain

1 o¢ OUz 1 o¢ ou
- - = -8(y- Yo)+-; - - = -8(x- x 0 ) - _z
G~ ~ G~ ~
)
Differentiating the first of these equations with respect toy and the second equation /
with respect to x and adding the results, we obtain

~(!._ o¢) + ~(!._ o¢) + 28 = o (14.17) Z X

ox G ox oy Goy
Figure 14.3 Establishment of boundary conditions
266 Introduction to the finite element method
FE formulation of torsion 267
as the x- and y-axes, respectively. We have
simplicity, we shall consider here situations where the member is solid, i.e. ( 14.26) is

n = [ :: J m = [ :: J InI = ImI = 1 ( 14.21)

valid.
The boundary condition ( 14.26) is then the one for which the differential equation
( 14.17) is to be solved. The objective of the analysis is primarily to determine the
The orthogonality of these two vectors is expressed through . she~!-~!~~~~~· an~.l!~a~nd once the 4>-function has been determined, these stresses
DTffi = 0; i.e. nxmx + nymy = 0 (14.22) are derived from-relation (14.16). For rather simple geometries of the cross-section,
like, for instance, ellipses and rectangles, exact closed-form solutions can be obtained
It appears that this relation is fulfilled for for 4> (cf. for instance Timoshenko and Goodier, 1970). For other more complicated

[::] = [ -::J (14.23)

geometries, it is obviou~--~hat we can obtain an FE formulation of ( 14.17) and ( 14.26)
similar to that oftwo-dimensional heat flow, and we shall investigate this topic later

This expression is also easily seen to fulfil the previously mentioned orientation of
the m-vector.
After these preliminary remarks, we are able to establish the boundary conditions.
- in etail.

14.1.3 Determination of torque

..

We recal~ t~e_torque T is applied at the ends of the member, i.e. the traction
vector t 1s~zero; along the boundary 2 of the cross-section. As the only non-zero In the previous section, we used the fact that the traction vector t is zero along the
stresses are ~and a,., and as n is located in the xy-plane, the boundary conditions prismatic boundary of the shaft to determine the boundary conditions for Prandtl's
(12.14) reduce to stress function 4>. Let us now consider the stress boundary conditions at the ends of
the shaft.
From (14.9) we have that a .. = 0 in accordance with the fact that only a torque
Inserting ( 14.16) and ( 14.23) yields Tis applied at the ends of the shaft. The remaining stress components axz and a,, at
the end of the shaft gjve...Iise to torque T and if these stress components are those
o4> o4> that follow from the solution of ( 14.17) then all field equations and boundary
- mx+ - m,= O
ax ay conditions are exactly fulfilled by our approach.
or One point needs some clarification. In the differential equation ( 14.17), the rate
of twist () enters. This rate of twist is most often unknown and instead the applied
(V4>lm=O 'f - -..t)j\2. .J:j ') ~ 0}~
.1. /l?"V'J ( 14.24) torque T is known. Let us therefore derive an expression for T. The torque T is
i.e. the gradient V4> is orthogonal to the tangential vector m. We have previously measured as positive if it acts in the counter-clockwise direction (see Figure 14.4).
shown in (5.8) that the slope d4> / dm of a quantity 4> in the direction of the unit vector Taking moments about an arbitrary point, for instance the origin of the coordinate
m is given by ~
'!>.'\'~
y
(14.25)

and a comparison with ( 14.24) shows that d4> I dm = 0, i.e. 4> is constant along the
boundary 2 of the cross-section. As the differential equation ( 14.17) only contains
derivatives of 4>, we may conveniently set
Yo
4> = 0 along boundary 2 (14.26)

This condition ,hold& if the cross-section of the member is simply connected, i.e. if z ~------~-----------x
the member is ~lid.)f the member is not solid, i.e. if the cross-section contains holes,
then 4> takes dif ent constant values on each individual closed boundary. For
Figure 14.4 Establishment of torque T
:J ~ 7
. ._.)
268 Introduction to the finite element method FE formul ation of torsion 269
system, we obtain from Figure 14.4 loading emerges only via the specification of the value of the rate of twist 8. If, instead,

T= L (xazy- ya.,J dA
the torque Tis known, we first solve ( 14.17) for an arbitrary value 8 1 of the rate of
twist. From the resulting ¢ -solution, the corresponding torque T1 is calculated
according to (14.28), and following (14.29) we then have
With (14.16) we obtain

T= -I (/J¢>
A OX
0
+y ¢>)dA=
Oy
-I A
(V¢>lxdA ( ~4.27)
8=!!..!_T
Tt
(14.31)

Finally, as the solution ¢> is proportional to the rate of twist (), the same holds for
where the shear stresses a xz and a yz.

X=[;] 14.1.4 Soap film analogy

Use of the Green-Gauss theorem (5.26) in (14.27) results in Let us combine the stress components ax. and a>'. into the matrix a given by

T = L ¢> div x dA- fy ¢xTn d..sf (1 4.32)

where the unit vector n is as shown in Figure 14.4. As we have chosen ¢> = 0 along Referring to Figure 14.4, it is obvious that a may be considered as a vector and the
the boundary Sf and as div x = 1 + 1 = 2, it follows that l(mgth of this vector is
( 14.33)
It appears that la l is the magnitude of the maximal shear stress occurring at a
particular position. Use of(14.16) in (14.33) gives
~t -~s, t~e to~q_ue i~. proportional to the volume beneath the surface described by ¢.
The solutwn ¢> of (14.17) subjected to the boundary condition (14.26) is
---------..
[(a¢)2 (a¢)2]1'2 =
= ox + oy
la l IV¢1 (14.34)
proportional to the rate of twist 8. Therefore, according to ( 14.28) the torque Tis
also proportional to () and we may write i.e. large shear stresses occur in regions where the gradient V¢ is large.
Equation ( 14.24) shows that V ¢> is orthogonal to the boundary. With n being
T = C() I ( 14.29) the outer unit vector normal to the boundary, we therefore have

w~ere C ;_s_t:~r-_!he S£e_9fic me er_;n questio~- Th;s constant;, d¢ = (V¢)Tn = IV¢11nl cos a= ± IV¢ 1
termed the torsional rigidlty ~r torsional stiffness.!ln most applications, the shear dn
modulus G( , is- a constant,' and in this case the differential equation reduces to since 1n 1= 1 and a = 0° or 180°. From ( 14.34) we then conclude that
the Poisson equation ( 14.18 ). This implies that the solution ¢> becomes proportional
to G(), i.e. instead of ( 14.29) we have
~ (14.35)
T = GKV() I (14.30) ~
holds along the boundary of the member.
where the proportionality constant Kv only depends on the geometry of the The result ( 14.34) opens the way for the appealing interpretation given by
cross-section. For a variety of different cross-sections, exact expressions for Kv may Prandtl' s soap film or membrane analogy. For a thin film of liquid, such as that of a
be found in the literature, for instance Roark ( 1975) and Timoshenko and Goodier soap bubble, the predominant force is the surface tension force S per unit length,
( 1970).
which may be considered as constant. Following Timoshenko and Goodier ( 197~
We mentioned previously that in the differential equation ( 14.17) or ( 14.20), the p. 304 ), the deflection w of a soap film loaded by a lateral pressure p turns out to be
. 270 Introduction to !he finite element method FE formulation of torsion 271

Soint-Venont theory VI osov t heory

d<P = 0
dn rlJ 0 lnnf
0 ri I
Figure 14.6 Applicability regions for Saint-Venant and Vlasov theory

of interest (differential equations of equilibrium, kinematic relations, constitutive

equations and boundary conditions) are satisfied exactly. We observe that, initially,
we did not try to solve all these equations directly, but rather guessed one part of
the solution, namely certain assumptions for the displacements and then determined
the remaining part of the solution rationally so that all equations of interest were
Figure 14.5 Illustration of c/>-function
satisfied. In general, such an approach is called a semi-inverse method and in the
present case one speaks of Saint-Venant's semi-inverse method.
The Saint- Venant theory is exact, but it should be noted that we have assumed
governed by the differential equation
that the warping u z given by (14.6), i.e. the displacement in the z-direction, can occur
o2 w o2 w p
-+ -+ - =0
without any constraints, since we have that (Jzz = 0 (cf. (14.9)). Often the warping is
ox oy s
2 2 (14.-36) restrained, for instance at the end surfaces of the member, and then deviations from
the Saint-Venant theory will occur. Generally speaking, it the warping is small, as it
This expression may be appreciated by referring to its one-dimensional counterpart - usually is for solid or closed thin-walled members, Saint-Venant theory provides close
the flexible string ( 4.28). However, the important point is that ( 14.36) is completely predictions. For open thin-walled members, however, the warping is frequently so
similar to (14.18 ). In addition, the boundary condition for ( 14.36) is clearly that large that Saint-Venant theory becomes questionable and in that case the more
w = 0 along the edges of the soap film. This boundary condition is in perfeat agreement complicated so-called Vlasov theory must be applied (see for instance Kollbrunner
with ( 14.26). Therefore, if a soap film is placed on a section with a similar geometry and Hajdin, 1972). However, the ramifications of this theory are .beyond the scope
as that of the cross-section of the twisted member and then loaded by a pressure, of this text. Following this discussion we may summarize the applicability regions
the deflection w is proportional to the function ¢. By investigating this soap film, it for Saint-Venant and Vlasov theory as shown in Figure 14.6.
bec_omes possible to identify very easily where large shear stresses occur, since these
reg1~ns - according to ( 14.34) - coincide with the regions where the slope of the soap
film IS large. Moreover, even in FE analysis, the imagination of such a soap film may
be helpful for a qualitative evaluation of the character ofthe ¢-function. An illustration 14.2 FE formulation
of the ¢-function is given in Figure 14.5 and it is seen that along a symmetry plane \

we have d¢ / dn = 0. From (14.20) and (14.26) the strong form of the torsion problem is given by

div(DV¢) + 28 = 0 inregionA ~-~\I )·'(l -h /],..~ (14.37)

¢ = 0 on boundary .9! ...__; (14.38)
14.1.5 Discussion of Saint- Venant torsion theory
where
We have previously discussed that if the torque Tis applied at the end surfaces of
the member in accordance with the shear stress distribution following from the
a¢]
ox
~- [~~
solution ¢ then an analytical solution of (14.17) subjected to (1 4.26) is the exact 0 1
- I· V - (14.39) -
solution to the elasticity problem considered and not just an engineering 1;a]-;; ·
approximation. That it is the exact solution follows from the fact that all equations
272 Introduction to the finite element method FE formulati on of torsion 273
The analogy with two-dimensional heat flow is striking ( cf. ( 6.32)- ( 6.34 )). Therefore, It may be somewhat surprising that we make this division of the boundary 2, since
we may obtain the FE formulation of the torsion problem by making suitable changes we emphasized that 4> = 0 holds all along the boundary (cf. ( 14.38)). However, when
in the FE formulation for two-dimensional heat flow. However, it is more instructive considering symmetry of a region, we do not know the </>-value along the symmetry
to obtain the FE formulation for torsion directly from ( 14.37). lines, but instead we know the value of d<J> jdn, namely that d</> / dn = 0 (cf.
Therefore, we first derive the weak form of ( 14.37). To achieve this form, multiply Figure 14.5), and this is the reason for ( 14.46). In the example considered later, we
( 14.37) by the arbitrary weight function v and integrate over the region A, i.e. shall see an application of this kind. However, along the boundary of the member

L [vdiv(DV</>) + 2v8] dA = 0 (14.40)

itself, we recall from (14.35) that d<J> jdn determines the maximum shear stress along
this boundary.
Having obtained the weak form we now introduce the approximation for </>. As
The first term is integrated by parts using the Green- Gauss theorem (5.26), i.e. usual we write

L v div(DV<J>) dA = fy v(DV</> )T n d2- L

(Vv)TDV</> dA
4> = Na ( 14.47)
where N is the global shape function matrix and a contains the </>-value at the nodal
Use of this expression in ( 14.40) provides the followingE eak fo-rffij points in the entire cross-section. That is,

L (Vv)TDV<J> dA = fy v(DV<J>)Tn d2 +2 L
vOdA (14.41)
4>1
rPz
(14.48)
N.]; a=
Let us evaluate the boundary term in more detail. From ( 14.39) it follows that
1
... -~
DV<J> = - V<J> "
where n is the number of~ points for the entire cross-sectio~ and a component
G
i.e.
N; depends on x andy, i.e. N; = N;(x, y). From ( 14.47) we obtam
(DV<J>)Tn = _!_(V</>)Tn ( 14.42)
G V</> = Ba where B = VN ( 14.49 )
From (5.8) we have that which implies that
d<J>
dn
= (V<J>)Tn ( 14.43)

whered</>/ dn is the slope of 4> in the direction ofn. Combining( 14.42) and ( 14.43) yields
B=
aN 1
ox
oNz
ox . . 0~"] ( 14.50)
[ oN 1 oN2 oN.
--
oy ay oy
(DV </> )T n = _!_ d<J> (14.44)
Gdn JInserting (14.49) into (14.45) gives
Using (14.44) the weak form (14.41) may be written as
. ( f (Vv)TDBdA)a=f v_!_d<J>d2+f v_!_d<J>d.!t'+2f vOdA
(Vv)TDV<J>dA=f v_!_d</>d2+f v_!_d</>d2+2f vOdA (14.45)
A y , G dn Y, G dn A

fA Y, G dn Y, G dn A
(14.51)

In this expression we have split the boundary term into the two parts The final step is to choose the arbitrary weight function v in accordance with
the Galerkin method. That is, we set
1 d<J> (14.52) •
known along .!t' h v = Nc
G dn
(14.46) Since v is arbitrary, the matrix c is arbitrary. From ( 14.52) it follows that
4>=0 known along 2 9
Vv =Be (14.53 )
274 Introduction to the finite element method
FE formulation of torsion 275

As v = vT, (14.52) may also be written as element load vectors, respectively (cf. (10.34) and (10.35)). Moreover, we can derive
the FE formulation of one element using only those degrees of freedom which belong
v = c TNT ( 14.54)
to this element. In the usual notation we obtain
Inserting (14.53) and (14.54) into (14.51) and using the fact that cis independent of
position result in
I K•a• = r· (14.60)

cT[(f BTDB dA)a -

A
f.!i',
NT_!_ d¢ d..<t'-
G dn
f ,
.!i'
NT_!_ d¢ d..<t'
G dn
where

- 2 J. NTe dA Jo=
f•=f~+fj ( 14.61)

and
As this expression should hold for arbitrary c T-matrices, we conclude that

( fA
BTDB dA)a = f .!i'
•
NT_!_ d¢ d..<t'
G dn
+ f
.!i'
,
NT_!_ d ¢ d..<t'
G dn
+ 2f NTO dA
A 1
(14.55)
/

which is the FE formulation for the torsion problem. The reader may notice that the ( 14.62)
derivations starting from the strong form ( 14.37) and ending with the FE formulation
( 14.55) contain a summary of much of what has been presented in previous chapters.
As usual we define the stiffness matrix K, the boundary vector fb and ~e load
vector f 1 by
e, the
K = f BTDB dA where D =_!_I
G
Having obtained the ¢-solution for a given rate of twist
calculated from (14.28), i.e.
torque T is

L
A

fb =
f
.!i',
T 1 d¢
N - - d..<t'
Gdn
+ f.!i',
T 1 d¢
N - - d..<t'
G dn
( 14.56)
T =2 ¢dA
( 14.63)

fl = 20 L NT dA
Using that ¢ = Na we may calculate the torque according to
( 14.64)

where it was used that e is constant (cf. (14.15)). We may then write (14.55) as
However, in this case it is also advantageous to calculate T from the contributions
I Ka = fb + f1 I (14.57) of each element. In ( 14.63 ) we can write the area integration as the sum of integrations
over all element regions Aa, i.e.
The force vector f is defined by
( 14.65)
I f = fb + fl (14.58)

which leads to the standard form where n. 1 is the total number of elements. In each element, we have ¢ = N•a•, i.e.

I Ka = f I (14.59) ( 14.66)

As in the heat flow problem, we may obtain the global stiffness matrix K and the
load vector f 1 as a summation of all expanded element stiffness matrices and expanded We finally mention that when the ¢ -solution has been o btained, then within each
276 Introd uction to the hnite el ement m ethod FE formula tion of torsion 277
element we have (14.29)- (14.31)). The local nodal point numbers i, j , k of the three-node triangles
start with i located at the asterisk shown in Figure 14.7(q) and, as usual, the sequence

V~ ~ V~ ~ mJ
i, j , k is taken in the counter-clockwise direction.
(14.67) First let us evaluate the element stiffness matrices. As the shear modulus G in
B' a'; the present case is constant, ( 14.62) yields

From this value of the gradient V¢ the shear stresses are determined according to
K• =.!. J s•Ts• dA
G A,
(14.16), i.e.

8¢ 8¢
I For the three-node triangle, the s•-matrix is constant (cf. (7.99)), i.e. we obtain

(Jxz =- ; (J = -O
- (14.68) (14.70)
oy yz X
Referring to Figure 14.7 (b), the element areas Aa are equal and given by
110 - 2 10 - 2 10- 4 (

14.2.1 Example A =----- =-- (14.71)

a 2 2 4 16
Consider the square shaft shown in Figure 14.7(a). As indicated we have four lines Inserting (14.69) and (14.71) into (14.70) gives
of symmetry, implying that only one-eighth of the total cross-section needs to be
K• = -i410-15BeTBe (14.72)
analyzed. Along these symmetry lines we have the boundary condition d¢ / dn = 0
as indicated in the figure. The fractional part of the cross-section is divided into two Next let us evaluate the element load vectors. From ( 14.62 ), we get
three-node triangles as shown in Figure 14.7 (b). This extremely simple mesh is chosen
in order to obtain . moderate hi:md calculations. The dimensions of the cross-section
ff = 2()f N•T dA
are given in Figure 14.7(b), and in addition we have A,

G =%X 10 11 N f m 2 ; () = 10 - 4 rad/ m (14.69) As the element shape functions for the three-node triangle vary in a manner similar
where the value for the rate of twist ()· is chosen arbitrarily ( cf. the discussion of to Figure 10.5, we obtain with ( 10.47) that

a)

~----~------~C
~
sym m e t ry a xes
'-....
b)
F.+{]
considered domain From (14.69) and (14.71) we then conclude that
• =0

(14.73)
A"'
J'-------lf-------"..? m etr y a xes
It is observed that this expression for the element load vector holds for all the elements.
After these preliminary calculations, e/emen.t 1 is considered. From Figure 14.7(b)
10' 2 10' 2
--m - m we have
I. ·I I· 4 + 4 ·I 10 - 2
Figure 14.7 (a) Square shaft: use of symmetry; (b) element mesh and global nodal numbers
X;= Yi = 0; Xk= Yk = - -
4
278 Introduction to the finite element method FE form ulation of torsion 279

That is, B• given by (7.99) becomes (14.76), we arrive at

0 - 1 0 1
B• = 200[ -
1
1 OJ 5 0 2 -2 0 10 - 8 2
-1 -1 2 K= -x 10-1 1 fl= - - (14.79)
8 - 1 -2 4 - 1 24 2
From (14.72) and (14.74) the element stiffness matrix becomes
0 0 - 1 1

K·=~ x 10-11[ ~ 0-1]

1 -1 (14.75)
These expressions are the final ones for the global stiffness matrix K and the global
load vector f 1•
Now let us evaluate the boundary vector fb. From (14.56) and Figure 14.7(a)
-1 - 1 2
we get
Referring to Figure 14.7(b), the local nodal numbers i, j , k correspond to the global
nodal numbers 1, 2, 3. That is, after the contributions from the first element _given
fb =·f NT.!_ d<f> d.'l' + f NT.!_ d<f> d.'l'
y .oc G dn y •c G dn
by( 14.75) and ( 14.73), the global stiffness matrix K and the global load vector f 1 become
and as d<f> f dn = 0 along ABC, it follows that

K = Kf• = ~ X 10 - I I
1
0
0 -1
- 1 0
0
10 - 8
f1 = Cff = - -
1

(14.76)
fb =f IL',c
NT.!_ d</> d.'l'
G dn
8 - 1 -1 2 0 24 1 Along the boundary AC the only non-zero global shape functions are N 2 and N 4• i.e.
0 0 0 0 0
0
Next consider element 2. From Figure 14.7(b) we have
(14.80)
10-2
X; = -2-, Yi = 0;
R4
That is, B• given by (7.99) becomes where the notations

(14.77)
. \ R2 =
!i'•c
f
N 2 .!_ d</> d;,; R4 =
G dn !i'•c
1 d</>
N 4 --d2
G dn f (14.81)

have been introduced for brevity.

From (14.79) and ( 14.80), the final FE system of equations is
From ( 14.72) and ( 14.77) the element stiffness matrix becomes
1 0 -1 0 <1>1 0 1
0 0 0 0
-1] 0 2 -2 0 0 R2 10 - 8 2
5

K' ~~ 10-"[ ~
0 - x10 - 11 +-- ( 14.82)
0 1 - 1 0
x
- 1 ; Ki· = -5 X 10- 11 8 - 1 -2 4 - 1 <1>3 0 24 2
8 0 -1 2 - 1 0 0 -1 0 R4 1
-1 -1 2
0 0 - 1 1 where the boundary condition </> = 0 along AC has been used, i.e.
( 14.78) <1>2 = <1>4 = 0 (14.83)
Referring to Figure 14.7(b ), the local nodal numbers i, j, k correspond to the global From the first and third equations of ( 14.82), we obtain
nodal numbers 2, 4, 3. For clarity, the expanded element stiffness matrix Ki• is also 8
given in (14.78). Therefore, adding the element stiffness components of (14.78) and ~ X 10- 11[ 1 - 1][<~>1] = 10 - [1]
the element load components (14.73) to the proper components of K and f 1 given by '8 • -1 4 <1>3 24 2
,
(
 FE formulati on of torsion 281
280 Introduction to the finite element method

with the solution follows from (14.67}, (14.74) and (14.85) that

(~~]~zoo[=:_: ~J[::J
2 10 3
¢1 = - x 10 3 = 133.3 N j m; ¢ 3 = - = 66.7 N f m (14.84)
15 15
In order to calculate the torque T and the stresses, it is advantageous to establish
the vector a• for the two elements. With reference to Figure 14.7(b) it follows that

element 1 a' ~ [ :J element 2 a'

To determine the torque T , we first evaluate the term

~ [ ::J (14.85)
and with the solution ( 14.83) and ( 14.84 ), we get

IA.
N· dA =I A.
[Ni Nj Nk] dA From (14.68 ), the shear stresses in element 1 then become
~
(Jxz = 0; (Jyz = 26.7 X
3
10 N / m
2 (14.90)
The element shape functions vary in the manner shown in Figure 10.5, i.e. (10.47)
implies that We emphasize that, due to the adoption of the three-node triangle, which possesses
a constant gradient V ¢, the stresses are also constant within the element. Proceeding

IA.
N•dA = Aa[1
3
1 1]

From (14.66}, (14.85) and (14.86) we obtain

.(14.86]-' in the same manner for element 2, we find that
( 14.91)

'
where the factor 8 emerges because we have only considered one-eighth of the total
cross-section. As A 1 = A 2 = Aa, where Aa is given by ( 14.71 ), and with the solution
(14.83) and (14.84), we arrive at
•
( 14.87)
This is the torque required to produce the assumed rate of twist () = l0- 4 [rad/ m]
(cf. (14.69)). The torsional rigidity C (cf. (14.29)) is
T 4
C = - = -10 3 = 88.9 N m 2 (14.88)
() 45
From Timoshenko and Goodier ( 1970, p. 313), the exact torsional rigidity can be
calculated as
Cexact = 112.5 N m 2 (14.89)
and, recognizing the extremely simple finite mesh adopted, the solution (14.88) is
surprisingly close.
Finally let us determine the shear stresses given by (14.68). For element 1, it
 IS _ _ _ _ ____ Approximating functions - vector problems

where only the linear terms in ( 15.1 ) have been considered. As all the parameters
283

{3 1 ... {34 and y 1 •. • y4 are independent, it appears from (15.2) that the
et 1 .. . et 4 ,

Approximating functions for I

approximation (15.1) for the displacement components implies that an arbitrary
constant strain. state can be approximated within a region. Moreover, as the terms
the FE method - vector a 1, {3 1 and y 1 correspond to an arbitrary rigid-body translation, we are led to the
following completeness requirements:
problems ,.
Completeness
• The approximation of the displacement vector u must be able to repre-
sent an arbitrary constant rigid-body motion.
• The approximation of the displacement vector u must be able to repre-
Up until now, the FE approach has been formulated for problems having one unknown sent an arbitrary constant strain state.
function. When dealing with solid mechanics we have, in general, more unknown
functions, namely the displacements in the x-, y- and z-directions. These displacements
define the displacement vector u and we shall now discuss the manner in which we · These completeness requirements are fulfilled for the approximations given by ( 15.1 ).
obtain approximate expressions for this vector. It will turn out that such In addition, we require fulfilment of the compatibility or conforming requirement
approximations are easily established when use is made of our previous results fr~ stating that :
Chapter 7. However, let us start with the convergence criteria.

Compata"b"l" ~J!- reqmremen

1 Jty or conJormmg
r · t

15.1 Converg~e criteria • The approximation of the displacement vector u must vary in a continuous
manner over element boundaries .

We consider a problem for which the unknowns are given in terms of the displacement •
vector u. For decreasing element size, it is required that the approximation fo( u If both the completeness and compatibility requirements are fulfilled, we have ensured
should converge towards the exact displacement vector. For an infinitely small region, convergence in the sense that, for a decreasing element size, the approximation for
the exact displacement components may either be constant or possess constant the displacement vector u approaches the exact one, i.e.
derivatives with respect to the coordinates, i.e. they exhibit constant gradients. In
three dimensions, the approximation for the displacement components must therefore • convergence criteria = completeness + compatibility requirements
be of the type
We observe that it may be allowable to relax the compatibility requirement provided
ux = et 1 + et 2 x + a 3 y + et4 z +possibly other terms that the discont.inuity across the element boundaries approaches zero for decreasing
uy = /3 1 + {3 2 x + {3 3 y + {34 z + possibly other terms ( 15.1) element size. We may refer to Hughes (1987), Strang and Fix (1973) and Zienkiewicz
and Taylor (1989) for a more stringent treatment. Such non-conforming elements
Uz = Y1 + y 2 x + y 3 y + y 4 z + possibly other terms are used es~cially in plate analysis, discussed in Chapter 18.
The cohtinuity requirement for the displacement vector u clearly means that the
where et 1 ... et4 , {3 1 .. . {34 and y1 ... y4 are constant parameters within the region displacement components are also required to vary continuously over the element
- or element - considered. With ( 15.1 ), we can determine the strains according to boundaries. With this observation we maycompare(15.1) with (7.2) and it is concluded
( 12.37) and ( 12.40) to obtain that each displacement component can be approximated in exactly the same manner
6xx = <X2; 6yy = /33 ; 6zz = Y4 as that used for temperature approximations. We may therefore refer to Chapter 7
( 15.2) for a discussion of the rate of convergence. Likewise, more information about
Yxy = et3 + {32; Yxz = et4 + Y2; Yyz = /34 + Y3
non-conforming elements, especially the patch test that such elements must pass, is

282
284 Introduction to the finite element method Approxima ting functions - vector p roblems 285

given on page 93. It is emphasized that the displacement components are approximated I y

~yi/7
independently of each other.

15.2 General approximation techniques

With the observation that every displacement component can be approximated in
~
exactly the same manner as that adopted for temperature approximations and l~odal point i
independently of each other, we can directly establish the approximation technique
for the displacement vector u.
X
First consider two-dimensional solid mechanics, where only the displacement
components ux and uy are considered. Assume that the element contains n. nodal Figure 15.1 Nodal displacements ux1 and uy 1 for nodal point i
points. For each nodal point i we have a displacement ux1 in the x-direction and a
displacement uy1 in the y-direction, as illustrated in Figure 15.1. From Chapter 7 it These two expressions can be combined into
follOWS directly that if the same type Of approximation is used for Ux and Uy, We get
I u = N•a• ( 15.5)

where
Ni ... N~J "(15.3)
uxl
Uyl

Nf belonging to nodal point i is

l
Ux2
It is emphasized that the element shape function
exactly the same as that discussed in Chapter 7. Moreover, as the same type of
approximation is used for ux and uy it follows that Nf in the expression for ux is the
u=[uxl N• =[N~0
Uy _
0
N•1
Ni
0
0
N•2
N •Pie
0
0
N•Pie '
a• = Uy2 (15.6)

same as Nf in the expression for uY. Instead of the formulation (15.3), we may write
Uxnc

Uyn,

In two-dimensional solid mechanics, we only consider the strains

ux = [N~ 0 Ni 0 . . . N~ OJ uy2 (15.7)

According to ( 12.44) and ( 12.45), these expressions can be written as

( 15.8)
( 15.4)
where I'
a
0
ox
Uy = [0 N~ 0 N2 ... 0 N~] Uy2
a ( 15.9)
0
oy
a a
oy ox
286 Introduction to the fin i te element method Approximating functions - vector problems 287
Combining (15.5) and (15.8) yields I where

~ ];
0 0 0 j Nn 0
e = B•a• = VN• [N, N z

·{J
where B• ( 15.10)
I I I N= ~ Nt 0 0 Nz 0 0 Nn
With (15.6) and (15.10) we get 0 0 0 0
0 Nt Nz Nn
oN~
- 0
oN;
- 0
oN~
0
ox ox ox u x!

B·= 0
oN~oN2 0 - 0
aN·n, (15.11)
Uy !
ay ay ay Uz!
aN1 -aN~ aN; aN; oN~ aN~ Uxz (1 5.1 6)
ay ox ay - ax ay ax a= Uyz

Defining a global shape function Ni belonging to nodal point i in exactly the Uzz
same manner as in Chapter 7 (cf. (7.135) and (7.136)), we can write the approximation
of the displacement vector u over the entire body as

(15.12L

where
and where n is the number of nodal points in the entire body. That is, N has the
dimension 3 x 3n and a has the dimension 3n x 1. From ( 15.15 ) and ( 12.42) we get

e = Ba where B = VN ( 15.17)
Nn
(15.13)
0 and

a 0 0
ax
a 0
and where n is the number of nodal points for the entire body. It follows that N has 0
Bxx oy
the dimension 2 x 2n and a the dimension 2n x 1. F rom ( 15.12) and with ( 15.9) we
obtain Byy a
0 " 0 oz
e zz
e= ( 15.18)
I e = Ba I where I B = VN (15.14) Yxy a a 0
Yxz oy ax
where B has the dimension 3 x 2n.
The generalization to three dimensions is straightforward and we obtain for the
Yyz a o a
approximation over the entire body that az OX
a a
0
u = Na ( 15.15) az oy
 I
288 Introduction to the finite element method Approximating functions- vector problems 289

The dimension for B is 6 x 3n. The approximation over one element follows the I According to (7.92) the element shape functions are given by
procedure above and we need not write the expressions explicitly.
From the discussion above it is evident that, by using the results of Chapter 7,
we may formulate a sequence of finite elements applicable to solid mechanics. As the
establishment of these elements is straightforward, we confine ourselves to !m
illustration of the two simplest two-dimensional elements: the three-node triangle (15.21 )
and the four-node rectangle.

15.2.1 Three-node element- constant strain element

In order to illustrate the approach outlined above, we consider the simplest possible where A is the a rea of the triangle. From (15.10) and (15.11) it follows that
element applicable for two-dimensional solid mechanics: the three-node trian~le s = B•a• (15.22)
shown in Figure 15.2. We recall that two unknowns are related to each nodal point:
a displacement in the x-direction and a displacement in the y-direction. As usual, where
nodal numbers i, j , k are taken in a counter-clockwise direction. From ( 15.5) and
(1 5.6) we have 1 [y,-y, 0 Yk- Y;
0
0
xk
Y;- Yi
0 x~x] ( 15.23)
u = N •a• (15.19)_ B• =-- 0 xk- xi X;- J I

2A
xk- xi Yi- Yk x;- xk Yk- Y; X j - X; Y;- Yi
where
uxi It appears that B• is a constant matrix, i.e. the strains s are constant within the
element. For this reason the element is often termed the constant strain element (CST
Uyi
element ).

~:l
N~ 0 N: U xj
J
a• = (15.20)
0 N~ 0 Uyj
J

uxk 15.2.2 Four-node rectangle - Melosh element

Uyk
As the next example consider the rectangular four-node element shown in Figure
' I
15.3- the so-called Melosh element; cf. Chapter 7. From (1 5.5 ) and (15.6) we have
,-
u = N •a• (15.24)
y

y t Uy4 t Uy3

u yi Uxj
'H::··J2b
~l_ux2
2
2a
I.
L-----------------------------x L------------------------X
Figure 15.2 Constant strain element Figure 15.3 Melosh element
290 Introduction to the finite element method Approximating functions - vector problems 291

where I B• consists of constant components, i.e. the strain e.x.x is constant. A similar observation
holds for a line parallel to the y-axis.
U.xl As in the discussion in Chapter 7, we recall that in o rder to fulfil the compatibility
requirement, the boundaries of the four-node element must be parallel to the
coordinate axes. We also mention that this restriction can be relaxed by means of
the isoparametric formulation presented in Chapter 19.

u=[::J N·=[~~ N~
0 N•2
0
0
N~
N•3
0
0
N•3
N•4
0
0 ]· a• =
N~ '

(15.25)
The element shape functions are given by (7.110), i.e.

N~ =
1
- - (x- x2)(y- Y4)
4ab

N •2 = 1
- - (x- x1)(y- Y3)
4ab ( 15.26)
1
N3 = - - (x- x 4 )(y - Y2)
4ab

With (15.10) and (15.11) it is concluded that

..
t = B•a• ( 15.27)

where

y - Y4 0 YJ- Y 0 Y- Y2 0
1 y,-y 0 ]
B• = - - 0 x- x 2 0 X1 -X 0 x- x 4 0 x3 - x
4ab [
x- x 2 y- Y4 XI- X Y3- y x - x 4 y - Y2 x 3 - x Y1- Y

(15.28.)
We o bserve that along a line parallel to the x-axis, i.e. y = constant, the first row of
 FE formulation of 3-D and 2-D elasticity 293

16 where
a a a 0 axx
0 0
FE formulation of three- and ox
a
ay oz
a 0 a
aYY
a,,
two-dimensional elasticity VT = 0
oy
0
ox oz a=
axy
a a a axz (16.2)
0 0 0
oz ox oy ayz

Carrying out the matrix multiplications of ( 16.1) gives (as a xy = aYX' etc.)

aaxx oaxy oaxz b

- + - + - + x= 0
One of the most important applications of the FE method is in three- and
ax oy az
two-dimensional elasticity. The approximations used for the displacements are those ( 16.3 )
treated in the previous chapter. In Chapter 13 it was emphasized that the expressions
obtained for the stresses and strains are general and - within the assumption of small
strains - they apply for any solid material. This implies that the only topic which aazx oa.y oazz b
-+-+-+ z = 0
differentiates elasticity theory from, for instance, plasticity theory is the form of the ox oy oz
constitutive relation. It is therefore of considerable importance to keep the formulation in accordance with ( 12.19). On the boundary of the body, the traction vector t with
as general as possible and only introduce the particular constitutive relation at the the components
latest possible stage. With such a viewpoint it becomes possible to formulate the FE
method easily, not only for linear elasticity, but also for more involved constitutive
( 16.4)
theories.
Since the differential equations of equilibrium hold irrespective of the constitutive
relation, we can achieve the objective discussed above by first making a weak
formulation of these differential equations. acts. Moreover, this traction vector must fulfil the boundary condition (12.14), i.e.

(16.5)
ty = ayxnx + ayyny + a,z nz
t. = azxnx + a.yny + azznz
16.1 Weak form of equilibrium equations- three-
dimensional case The objective is to determine the weak form of the differential equations of equilibrium.
Before doing so, we shall first derive a preliminary result.
For three-dimensional problems, the differential equations of equilibrium are given Consider the arbitrary vector v
by (12.21), i.e.

( 16.6)
( 16:1)

292
294 Introduction to the finite element method
FE form ulation of 3-D and 2-D elasticity 295
In accordance with the kinematic relation ( 12.42), we have
In a similar way, the second and third equations of ( 16.3 ) are multiplied by the
avx arbitrary functions vY and Vz respectively, and we find that
ax
( 16.10)
avy
ay
( 16.11)
OVz
az
Vv = ( 16.7)
avx OVy
- +-
oy ax
OVx OVz
-oz +-
OVy
ox
av.
-
l[
V
-ovx (J XX
~
+ -ovy (J yy + -ovz (J + (ovx
~ &
ZZ -;- + -ovy) (J + (ovx
~ 0X
XY - + -ovz) (J
0Z ~
XZ

-oz +-
ay + (OVy + OVz)a
oz oy yz
JdV = O
i.e.
Finally, use of (16.8) provides
(v )T
V (1 =
avx (J
-0 XX + OVy
- (1yy + -avz (J + (av
ZZ
av )
~ + _Y (J
x ay az ay ax xy ( 16.12)
( 16.8)
+ ( ~x + ~~) (J:x:z + ( ~; + ~~) (J yz
This is the weak form of the differential equations of equilibrium ( 16.1) subjected to
"W_e are now in a position to derive the weak form of (16.3). Multiply the first the boundary conditions ( 16.5), and we stress that the weight vector v is arbitrary
equa:wn of ( 16.3) by the arbitrary function vx and integrate over the volume V to and that ( 16.12) holds for any constitutive relation. This weak form is of fundamental
obtam importance in solid mechanics where it is often termed the virtual work equation or
virtual work principle. It is emphasized that the arbitrary weight vector v and the
oaxx dV + J Vx~dV
oa J oa dV + J vxbx dV = 0 stresses a are completely unrelated, and even though v may be interpreted as a
Jvvx-a
x v ay + vvx---==
az v displacement vector, v bas nothing to do with the real displacements of the body.
Clearly, v may be chosen as the real displacements u, but in general this is not the
A~ integration by parts using the Green- Gauss theorem is now performed. Using case. When the weak form ( 16.12) is derived, the traction vector t emerges in the
thts theorem in the form given by ( 5.34 ), the result becomes boundary term. In accordance with the discussion of heat flow, the traction vector

f ~vx (Jxx dV + f Vx (Jxyny dS - f avx (Jxy dV

t is therefore a natural boundary condition.
f Vx 11xxnx dS-
The celebrated virtual work principle was derived from the differential equations
Js Jv ux Js JvaY of equilibrium and the boundary conditions for the traction vector. Just as for the
heat flow problem, it can be shown that as v in the weak form (16.12) denotes any
+ f Vx(Jxznz dS- f ovx (Jxz dV + f vxbx dV = 0
Js Jv oz Jv vector, the weak form implies ( 16.1 ) and ( 16.5). The reader is referred to Hughes
( 1987) for details.
The component tx of the traction vector tis given by ( 16.5), i.e. the expression above
reduces to
16.2 FE formulation of three-dimensional elasticity
06.9) With the weak form ( 16.12) of the equilibrium equations, it is straightforward to
derive the FE equations for three-dimensional elasticity. As mentioned previously,
296 Introduction to the finite element method
FE formulation of 3-D and 2-D elasticity 297

we shall introduce the assumption of elasticity at the latest possible stage. To achieve ( 12.42) states that
this purpose, we already know that the displacement vector u will be approximated by ( 16.18)
e = Vu
u=Na (16.13)
From (16.13) and (16.15) we get
(cf. (15.15)). The Galerkin method means that the weight vector v is chosen in ( 16.19)
e = Ba
accordance with
i.e. (16.17) becomes
v = Nc ( 16.14)
11 = DBa - De (16.20)
0
As v is arbitrary, the matrix c is arbitrary. From ( 16.14) it follows that
With (16.20), (16.16) takes the form
Vv = Be where B = VN ( 16.15)
(LBTDBdV)a= tNTtdS + LNTbdV+ LBTDe 0 dV (16.21)
(cf. (15.17)). Inserting (16.14) and (16.15) in the weak form (16.12), and noting that
c is independent of the coordinates, yields Let us now consider the boundary conditions, which are expressed either in terms
of a prescribed traction vector t - the natural boundary condition - or a ~rescribed
displacement vector u - the essential boundary condition. Instead of the tract10n ~e~tor
given by (16.5), we may use the formulation (12.13) and the boundary conditions
As the c-matrix is arbitrary, we conclude that become

t = Sn = h on Sh (16.22)
(16.16)
u =g on Sg

As this expression was derived from the weak form of the equilibrium conditions where h and g are known vectors. This means that the traction vector t is known
that hold for arbitrary weight vectors v, ( 16.16) holds exactly and expresses the along the boundary Shand the displacement u is known along the boundary Sg. The
balance principle for the body. It is emphasized that ( 16.16) applies irrespective of total boundary S consists of Shand S9 (note the difference between the total boundary
the constitutive model. Sand the stress tensorS). With these formulations (16.21) takes the form
The global shape function matrix N has the dimension 3 x 3n, where n is the
number of nodal points for the entire body (cf. (15.16)), and as b has the dimension
3 x 1, NTb has the dimension 3n x 1. This means that (16.16) constitutes a system (fv
BTDB dv)a =I
s.
NTh dS
s.
+I
NTt dS +
v
f ~bdV +Iv
BTDeo dV
(16.23)
of equations with 3n equations. As in the discussion of heat flow ( cf. Chapter 10) the which is the FE formulation sought.
right-hand side of ( 16.16) can be viewed as forces acting at the nodal points. These In order to write this formulation in a compact fashion, the following matrices are
nodal forces have components in the x-, y- and z-directions and, as in the discussion defined:
in Chapter 10, it can be shown that the nodal forces fulfil exactly the conditions for
global equilibrium of the entire body. This is certainly not surprising since {16.16)
was derived from the equilibrium conditions for the entire body. K = L BTDBdV

I I~
At this stage we introduce the constitutive model, and we assume that the material
responds thermoelastically. Consequently, the constitutive model
fb = NTh dS + t dS
s. s. ( 16.24)
I = De - Da0 I (16.17)
r. = L~bdV
11

is adopted ( cf. ( 13.23), where D is the constitutive matrix and a0 contains the initial
strains. These initial strains are assumed to be known. In most cases, a0 is due to
thermal strains, in which case it is determined by an independent temperature f 0 = L BTDe0 dV
calculation, for instance in terms of a heat flow FE calculation. The kinematic relation
298 Introduction to the finite element method FE form ulation of 3-D and 2-D elasticity 299

where K is the stiffness matrix, fb the boundary vector, f1 the load vector and f0 the for all a ; only if a corresponds to a rigid-body motion is ~he quadratic form above
initial strain vector. Using (16.24), (16.23) can be written as equal to zero. These observations are in complete an.a~ogy w1th the he~t flow problems,
and it follows that if sufficient boundary conditions for the displacements are
I Ka = fb + f1+ f0 (16.25) prescribed so that rigid-body motions are elimi~ated, then the corresponding
submatrix K of K is non-singular. In analogy with the heat flow problem, the
If n is the number of nodal points for the entire body, it can easily be seen that K prescription-of these boundary conditions allows for a modification of the system of
has the dimension 3n x 3n, a has the dimension 3n x 1 and the right-hand side of equations (16.27) so that a unique solution can be o btained. . .
(16.25) has the dimension 3n x 1. Moreover, defining the force vector f by As before, we may obtain the FE equations for one element, and w1th obvious
notation we get
1 f = fb + f1 + fo 1 ( 16.26)
(16.31)
we obtain the standard FE formulation
where

I Ka =f I ( 16.27) (16.32)

where f has the dimension of force, i.e. [N]. We observe that the inclusion of initial
strains £ 0 in the analysis presents no difficulties in the FE formulation, and it is and
characteristic for the FE method that various effects may easily be accounted for.
We note that in order to determine the initial strain vector f 0 , it is required that the
initial strains e0 are known. If these initial strains are thermal strains, then an FE
analysis of the corresponding heat flow problem is first carried out. With the known "
temperatures, the initial strain vector t 0 is determined from ( 13.24 ), and with ( 16.25)
we obtain the elastic response of the body exposed to combined mechanical and
f
fb =
s..
NeTh dS + f
s..
NeTt dS
(16.33)
thermal loading.
The contribution of the load vector f1 from concentrated forces can be dealt with
~=f NeTbdV
v.
by means of Dirac's delta function in complete analogy with the treatment of point
sources in heat flow problems.
According to ( 13.4 ), D is symmetric; that is, it follows from ( 16.24) that the
stiffness matrix K is also symmetric, i.e.
In these expressions, Va and Sa are the volume and surface of the element, respectively.
In general, the boundary conditions along the element surface are unknown.
(16.28)

We observe that if a corresponds to a rigid-body motion, this creates no strains in

the body, i.e. t = Ba = 0. As in the discussion of (9.38)- (9.40) it follows that K is
singular, i.e. 16.3 Weak form of equilibrium equations - two-
dimensional case
I det K =0 ( 16.29)
In the present text, two-dimensional solid mechanics is defined according to:
Moreover, as D is positive definite ( cf. ( 13.11 )), it follows from ( 16.24) that K is ( 16.34)
two-dimensional problem = plane stress or plane strain conditions
positive semi-definite, i.e.
In general, however, ax isymmetric problems can also be treated as a two-dimensio~al
( 16.30) problem and even though the FE formulation of such problems can be dealt with
 FE formulation of 3-D and 2-D elasticity 301
300 Introduction to the finite elemen t method

in a fashion closely related to the exposition given below, we shall confine ourselves y
a) b)
here to the definition given by ( 16.34 ).
For plane stress conditions, we may derive the weak form of the equilibrium
z
differential equations ( 12.26) in a manner completely analogous to the derivation of
the weak form ( 16.12) applicable to three-dimensional solid mechanics. However, we
want to obtain the weak form of the equilibrium differential equations not only for
plane stress, but also for plane strain conditions, and we shall perform this derivation
in a unified manner here. L---------X
lL_,
The weak form ( 16.12) of the equilibrium equations holds for the general Figure 16.1 (a) Region A and boundary fL' for two-dimension~\ problems; (b) illustration
three-dimensional case, so it also applies for two-dimensional problems. In ( 16.12), of upper surface S +, lower surface S- and thtckness t
the weight vector vis arbitrary and it is allowable to choose its components according
to
V:x: = Vx(X, y); Vy = Vy(X, y); Vz = 0 ( 16.35) For the two-dimensional problem described by the x- and y-coordinates, we d~note
This choice implies that (16.8) reduces to the region in the xy-plane by A and the boundary of t.his r~gion by !£'; see Ftgure
16.1 (a). The thickness t in the z-direction is s~own m F1gure 16.1 \b) (note t~e
- T ovx OVy (OV;r OVy) difference between the thickness t and the traction vector t). As all mtegrands m
(Vv) C1 = - (Jxx + - (JYY + - + - (Jxy ( 16.36)
ox oy oy ax (16.39) are independent on the z-coordinate, we obtain
From the following redefinition of matrices
a
t (Vvfa d v = L[f. J L (Vv)Ta dz dA = (Vv)Tat dA
(16.40)
ox
0
tvTb d v = L[f. J L vTb dz dA = vTbt dA
V= 0

a a
a
oy v=[~:J [""]
a = (JYY
(J xy
(16.37)
Consider now the surface integral in ( 16.39). Referring to Figure 16.1 (b), lets+ and
s- denote the upper and lower surface of the body, respectively. We then get that
oy ox
it follows that (16.36) is fulfilled. We note that this definition of the operator V occurs f s
vTt dS = f
s+
vTt dS + f
s-
vTt dS +
rl
Sf
[f VTt dz Jd.sf
z
(16.41)

also for plane stress (cf. (12.27)) as well as for plane strain (cf. (12.44)). We also Referring to (16.38), the traction vector tis given by the components tx and ty. Along
introduce the definitions s+ and s- we have nx = ny = 0, i.e. (16.5) gives

t = [::l [::J
b = (16.38)
t= [::J
= [:::::]
With the definitions ( 16.37) and ( 16.38), the weak form ( 16.12) becomes along s+ and s-. For plane stress conditions, we have by definition that (Jxz = C1yz = 0;

t(Vv)TadV = t vTtdS + tvTbdV (16.39)

that is, (16.41) reduces to

t f vTt dS = Sf VTtt d!£' ( 16.42)

This formulation looks like (16.12) but it is different since v, V, a, t and b are now
defined by (16.37) and (16.38). For plane strain conditions, the components (Jxz = (Jxz(x, y) and (Jyz =-= C1y•Sx, Y) may,
Let us now observe a characteristic feature of two-dimensional problems, namely: in general, be different from zero. However, along the upper. surfac~ S we h.ave
n = 1 whereas n = - 1 holds along the lower surface s-. W1th obvwus notatwn,
For two-dimensional problems, the displacements, strains, stresses, tractions this i~plies that t ~ = -t - and as vis independent of the z-coordinate, ( 16.41) reduces
and body forces do not depend on the z-coordinate. to ( 16.42) even for plane strain conditions.
r

302 Introduction to the finite element method FE form ulation of 3-D and 2-D elasticity 303

Therefore, insertion of (16.40) and (16.42) in (16.39) yields and as the c-matrix is arbitrary, we conclude that

( 16.43) ( 16.49)

which is the weak form~ or the virtual work principle ~ for two-dimensional problems. The similarity with (16.16) should be noted. As ( 16.49) is only based on the equilibrium
Again we recall the difference between the thickness t and the traction vector t. In equations it holds for arbitrary constitutive models.
( 16.43 ), the thickness t may vary, but in order to approximate the requirements for We now introduce our assumption for the material behaviour, i.e. the constitutive
the existence of plane stress (or plane strain), such a variation must, in general, be relation. In the present case we assume thermoelasticity and we note from ( 13.32 ),
small. It is emphasized that ( 16.43) holds irrespective of the constitutive model. applicable for plane stress, and from (13.39), applicable for plane strain, that both
If the thickness t of the body is constant, the virtual work principle may be cases can be formulated as
simplified to

L (Vv)Ta dA = f2' vTt d!t' + L vTb dA (16.44)

I a= De - Da 0
( 16.50)

where

a= (jaxx]
,, ; a =
["xx]
e,, ( 16.51 )
[ (jxy 'Y xy
It is emphasized, however, that D and e0 take different forms depending on whether
16.4 FE formulation of two-dimensional elasticity plane stress or plane strain is assumed: cf. (13.33) and (13.30) with (13.38). Just as
for three-dimensional elasticity, the initial strains a0 are assumed to be known. For
The derivation of the FE formulation from ( 16.43) closely follows that of thermoelasticity, e0 is due to temperature changes which are determined by an
three-dimensional elasticity. As before, we introduce the specific assumption for the independent temperature calculation, for instance in terms of a heat flow FE analysis.
constitutive relation at the latest possible stage. We already know that the From (16.50) and (16.51) it follows that the stresses given by a only depend on
displacement vector the in-plane strain components Bxx• en and Yxy• and this is the reason why the
displacement vector u given by ( 16.45) only comprises the components ux and u,
u = [::] (16.45) even though uz for plane stress conditions, in general, is different from zero.
Irrespective of whether plane stress or plane strain is assumed, the strains e
is to be approximated by defined by ( 16.51) are derived from the displacements ( 16.45) using the kinematic
relation
u = Na ( 16.46)
a=Vu (16.52)
The reason that u only comprises ux and u, and not uz will be revealed in a moment.
According to the Galerkin method we choose the arbitrary weight vector v as From (16.46) and (16.48) we obtain
v = Nc (16.47) e=Ba ( 16.53)

As v is arbitrary, c is also arbitrary. From ( 16.47) it follows that i.e. (16.50) takes the form
Vv = Be where B = VN ( 16.48) a= DBa- Da0 ( 16.54)

(cf. (15.14)). Inserting (16.48) in the weak form (16.43) gives Inserting ( 16.54) into ( 16.49) results in

cT(L BTat dA - f 2
NTtt d!t'- L~bt dA) = 0 (L BTDBt dA )a = f 2
NTtt d!t' + L~bt LB~e0 t
dA + dA ( 16.55)
304 Introduction to the finite element method FE formulation of 3-D and 2-D elasticity 305
~he unit vector n normal to the boundary !l' is located in the xy-plane and the u =g
y
tract1on vector t along !l' is given by ( 16.38). That is, the traction boundary conditions
( 16.5) can be written as
t = Sn (16.56)
t =h
where
n
( 16.57)
L----------------------x
All boundary conditions can therefore be given in the form Figure 16.2 Two-dimensional region with boundary .? = .?h + .?9

initial strain vector. Using (16.60), (16.59) can be written as

t = Sn = h on !lJ.
( 16.58)
u=g on Zg I Ka = fb + r. + fo (16.61)

where h and g are known vectors. This means that the traction vector t - the natur~l If n is the number of nodal points for the entire body, it is easily shown that K has
boundary condition - is known along the boundary !lJ. and the displacement vector the dimension 2n x 2n, a has the dimension 2n x 1 and the right-hand side of ( 16.61)
u - the essential boundary condition - is known along the boundary !l'.. The total has the dimension 2n x 1. As usual, the force vector f is defined by
boundary !l' consists of !lJ. and Zg; see Figure 16.2. Using ( 16.58) we may ~eformulate
(16.55) according to 1 r = rb + r. + fo (16.62)

(f BTDBt dA)a = f NTht d!l' + f NTtt d!l' + f NTbt dA That is, ( 16.61) takes the standard form
~ Sf.
A A
I I
+ L BTDe0 tdA ( 16.59)
Ka =f ( 16.63)

where it can easily be seen that f has the dimension of force (i.e. [N]) just as for the
three-dimensional case (cf. ( 16.27)). Indeed, the close analogy with the formulation
which is the FE formulation for two-dimensional elasticity. for three-dimensional elasticity is distinct. Even the analogy with one-dimensional
I.n order to write this formulation in a more compact fashion the following elasticity, i.e. the response of an axially loaded elastic bar (cf. (9.133) and (9.131)),
matnces are· defined: should be noted. It is recalled that in the formulation presented for two-dimensional
problems, the thickness t may vary, but in order to approximate the requirements
K = LBTDBtdA
for the existence of plane stress or plane strain, such a thickness variation must, in
general, be small.

fb = f ~ht f
~
d!l' +
~
NTtt d!l'
As before, the contribution of the load vector f 1 from concentrated forces can be
dealt with by means of Dirac's delta function in complete analogy with the treatment
of point sources in heat flow problems.

r. = LNTbtdA
( 16.60)
We have seen that plime stress and plane strain problems result in FE
formulations which, in principle, are identical. As already. mentioned, they differ only
in the expressions for the constitutive matrix D and the initial strains e0 . The reader's
fo = LBTDe 0 t dA
attention is also drawn to the fact that even axisymmetric problems can be dealt with
in a fashion very similar to that of plane stress and strain; see e.g. Zienlciewicz and
Taylor ( 1989).
We also mention that for both two- and three-dimensional elasticity, the B-matrix
where K is the stiffness matrix, fb the boundary vector, f 1 the load vector and f 0 the contains first order derivatives of the global shape functions. With reference to the
306 Introduction to the finite element method FE formulation of 3-D and 2-D elasticity 307

discussion in Chapter 9, we conclude that C0 -continuity is required for the global y

shape functions. This C0 -continuity requirement implies fulfilment of the compatibility
4~------r-----~3
requirement discussed in the previous chapter. For more details of the FE approach
for two- and three-dimensional solid mechanics problems in elasticity, plasticity and
creep we may refer, for instance, to Bathe (1982), Cook et al. (1989), Hughes (1987), b-~~ 2b
X
Owen and Hinton ( 1980) and Zienkiewicz and Taylor ( 1989). thickness I
Finally, let us obtain the FE formulation for one element. With evident notation, =constant
the result is
2

K•a• = f" ( 16.64) 2o

J
I· ·I
where Figure 16.3 Four-node rectangle - Melosh element

( 16.65).
y-axis, i.e.
and
b- [ 0
-pg
J (16.67)

K• = f B•ToB•t dA where p is the mass density and g the acceleration due to gravity. From ( 16.66 ) it
A, follows that
N•1 0 0
( 16.66) 0 Ni N•1
N~ 0 0

ff = f
Aa:
N•Tbt dA = - t f Aa
0
N'3
N~

0
J
[ O dA = - pgt
pg
f
A.
N2
0
dA

0 N'3 N'3
N~ 0 0
0 N~ N~

(16.68)
16.4.1 Example 1

As the FE formulation fo r two- and three-dimensional elasticity contains two and where the assumption of a constant thickness twas utilized. From ( 15.26) and Figure
three unknowns per nodal point, respectively, it becomes difficult to illustrate the 16.3, we have
FE solution process by moderate hand calculations. However, in order to provide
some kind of illustration, the element load vector ff will be considered in this example
and the boundary vector fb in the next example.
The four-node rectangle shown in Figure 16.3 is adopted and the problem is to
f A.
NidA= -
1
4ab
fb (fa
-b -a
(x-a)(y-b)dx)d y = ab (16.69)

determine the element load vector f1' due to the body forces b. In the present example, and with the geometrical interpretation of the element shape functions in mind, it is
these body forces are due to gravity and they act in the negative direction of the evident that the same value is obtained for the other integrations. Using ( 16.69) in
308 Introduction to the finite element method FE formulation of 3-D and 2-D elasticity 309
(16.68) we obtain
0
a) 2
qq , -
_(_ IIIII [[l
1 A
---...---- B
4 3

r. = - pgabt
0
1
(16.70)
2b YL X
0
2

2o
0
I· ·I
Figure 16.5 (a) Local coordinate system and uniform traction; (b) equivalent nodal forces
It is evident that pg4abt is the total vertical force acting on the element. The result
(16.70) shows that one-fourth of this total force is distributed to each nodal point. tractions to the nodal forces is given by

16.4.2 Example 2 I
~E
NTht d2 = I
.si'Aa
NTht d2 + I
Sl'ac
NTht d.P + I
.5/Co
NTht d.P

Assume that a rectangular disk is loaded by a constant traction vector along the (16.72)
boundary AE; see Figure 16.4. The forces q1 and q 2 per unit area act in the directions
shown, i.e. the traction vector t along AE is given by '
As in the assembling process, we may evaluate these contributions using the element

~:J = [ -::J
shape functions and then - using the topology data - determine the proper locations
( 16.71)
t = h= [ in the boundary vector for these contributions. Along AB, we therefore consider the
term
In accordance with ( 16.58) the notation h is used for the traction vector, since it is
a known quantity. As in the heat flow problem (cf. the discussion of(10.29)- (10.33))
it is evident that the tractions along AE are equivalent to some point forces that act
at the nodal points. Our objective is to determine these equivalent nodal forces.
To evaluate this expression we may actually use any convenient local coordinate
Referring to (16.60) for the boundary vector fb, the contribution from the known
system and we choose the one shown in Figure 16.5(a).
Along AB the only non-zero element shape functions are N; and N~ and from
!_1 11
_ 11
_1_1 1_
11_1 I_ 1 1_.--q
11--q, 2 (15.25) and (16.71) we obtain fo r a constant thickness t
N1 0 0
A B c 0 E 0 N•1 0
Ni 0 0

2a 2b I SI'AB
N•Tht d2 = t I~B
0
N;
Ni
0
[ _ ql
q2
Jd.P = t I ~.
0
qtN;
d2

0 N•3 - q2N;
N~ 0 qlN~
0 N •4 -q2N~
Figure 16.4 Boundary with constant traction vector; four-node rectangles of the same size
310 Introduction to the finite element method

17
[ FE formulation of beams
I
2o
2b

-
The FE formulation just presented for two- and three-dimensional elasticity theory
Figure 16.6 Equivalent nodal forces for the loading shown in Figure 16.4
was based on the exact field equations (equilibrium, kinematics and constitutive
relation) and the corresponding boundary conditions. In many cases, however, the
As both N'J and Nl vary linearly along AB, we derive, for instance, that body is of such a geometry that it becomes possible to introduce a sequence o£

t f Z...e
q 1 N'3 d.P = tq 1 f
Z...e
N'J d.sf = tq 1 x !2a = q 1 at
assumptions which significantly simplify the problem formulation. Two of the most
outstanding examples of such an approach are the theories for beam and pla.te
bending. In principle, such structures are three dimensional, but as a beam is
I.e. dominated by its extension in the axial direction and a plate by its extension in the
0 plane, it becomes possible to make certain assumptions about the structural
deformation (i.e. about the kinematic relations), which significantly simplify the
0 problem. However, one should be aware of the fact that, apart from very special
0 situations, these theories are engineering approximations that violate some of the
field equations.
0
(16.73) A beam with cross-sectional area A(x) is shown in Figure 17.1. The x-axis is
q 1 at directed in the axial direction and the transverse loading q(x) is measured as positive
- q 2 at in the z-direction. This loading q is the force per unit length, i.e. q has the dimension
[N/ m]. The material and cross-section of the beam are assumed to be symmetric
q 1 at about the xz-plane, as also indicated in Figure 17.1. Moreover, only loadings normal
- q 2 at to the xy-plane and located symmetrically about the xz-plane are considered. This
implies that the deflection w of the beam occurs in the xz-plane and w is measured .
That is, the uniformly distributed traction shown in Figure 16.5(a) is equivalent to as positive in the z-direction. For beam theories considering non-symmetri~
the nodal forces shown in Figure 16.5(b ). It is obvious that the same results are
obtained when considering the other elements and this implies that the uniform
traction along A, B, C, D and E of Figure 16.4 is equivalent to the nodal forces z.w
shown in Figure 16.6.

a b

Figure 17.1 Configuration and loading of beam

311
312 Introduction to the hnite element method FE fo rmula tion of beams 313

cross-sections and more general loadings, we may refer to Timoshenko and Gere q dx

.rr rt
ZL
(1972).
tt

17.1 Beam theory

17.1.1 Equilibrium conditions
y X

~! D l~v·'" dx

For a section normal to the x-axis, we have the stress components (Jxx• (Jxy and (Jxz· Figure 17.3 Infinitely small part of beam
Due to the symmetry in geometry and loading about the xz-plane, we can restrict
ourselves to consider the effect of the components (Jxx and (Jxz for which we have

( 17.1) For an infinitely small part of the beam, the remaining forces and moments caused
by (J xx and (Jxz are shown in Figure 17.3. Vertical equilibrium requires that
These stress components give rise to a bending moment M and a vertical shear force
V defined by
q dx - V + (V + d V) = 0

i.e.

M = L Z(JXX dA ; v= L (Jxz dA ( 17.2)

( 17.5)

where M is the moment a bout the y-axis. In accordance with the positive directions
for (Jxx and (J xz ( cf. Figure 12.2 ), the positive directions for M and V are shown in Evaluating the moment equilibrium, for instance, about the left end of the infinitely
Figure 17.2. small part shown in F)gure 17.3, we obtain
Let us now establish the equilibrium conditions for the beam. First it is noted dx
that the stress component (J xx also results in a norma/force N in the x-direction given ~y M + q dx - + (V + dV) dx- (M + dM) = 0
2
i.e. ·
(17.3)
1 dM
- q dx + V + dV - - = 0
2 dx
As no resulting forces act in the x-direction, horizontal equilibrium implies that As q dx and dV are infinitesimal quantities, we conclude that
(17.4)
~
N=O
( 17.6)
~

17 .1 .2 Kinematic relations

The essential feature of beam theory is that certain kinematic assumptions are made.
Figure 17.2 Positive directions for bending moment M and shear force V The fundamental kinematic assumption for engineering beam theory, which dates
314 Introduction to the finite element method
FE formu lation of beams 315

Byy = Bzz = Y xy = Y yz ~ ")' xz =0 (17.11)

i.e. the only non-zero strain component is exx·

17.1.3 Constitutive relation

w
Linear elasticity in terms of Hooke's law for isotropic materials is assumed. According
to (17.10) and (17.11) the only non-zero strain is exx> i.e. use of(13.2) and (13.17)
results in

(17.12)
p

Figure 17.4 Deformation of Bernoulli beam and

(Jxy = (Jxz = (Jyz =0 (17.13)
back to Jacob Bernoulli (1654-1705), is expressed by:
A comparison of (17.13) with (17.1) reveals evident contradictions, since (17.13)
Plane sections normal to the beam axis remain plane and predicts the shear ·stress crxz to be zero even though in reality it must be non-zero in
normal to the beam axis during the deformation. order to have a non-zero shear force V. Stated explicitly, we have crxz "# 0 and Y xz = 0
and the constitutive relation cr xz = G y xz· This kind of inconsistency is typical of
To investigate the consequences of Bernoulli's assumption, a plane section normal to simplifying engineering theories which try to reformulate problems that are actually
the x-axis before the deformation is considered (see Figure 17.4 ). The two points P three dimensional into a simpler form.
and Q are located on this plane and point Q is located on the x-axis. Due to·the Instead of the equations above, a uniaxial state of stress is assumed; that is, we
deformation, the points P and Q move to the positions P ' and Q ' and, following assume that the relation
Bernoulli's assumption, the plane defined by P'Q' is normal to the deformed beam
axis. Since P is located below the x-axis, the distance (a positive quantity) between crxx = Eexx (17.14)
P and Q and between P' and Q' is given by - z. From Figure 17.4 and assuming
that the slope dw /dx is small, the displacements ux and u. in the x- and z-directions holds and we accept the contradiction (Jxz "# 0 and Y x z = 0. We may even include the
of point P are given by effect of initial strains (e.g. thermal strains) as
dw (17.15)
u = u0 - z-· u = w (17.7)
x dx' z
but for simplicity we shall not pursue this topic in wha t follows.
0
where u is the displacement in the x-direction of the point Q. Moreover, it is assumed
that 17.1.4 Choice of position of x-axis
Uy = 0 ( 17.8)
Insertion of (17.10) into (17.14) yields
and that u 0 and the deflection w only depend on x, i.e.
du 0 d2 w
u 0
=u 0
(x); w = w(x) (Jxx = E- - Ez - -2 (17.16)
( 17.9) dx dx
With these kinematic assumptions, the strains given by ( 12.37) and ( 12.40) become and using ( 17.2) the bending moment M is determined from
du 0
d2w
e = - - z--2 (17.10) (17.17)
xx dx dx
316 Introducti on to the finite element method FE formulation of beams ' 317

where u 0 and w only depend on the coordinate x (cf. (17.9)). Up until now we have gives
not specified the location of the x-axis - we have just said that the x-axis should be
du 0
in the axial direction of the beam. To obtain the simplest possible formulation, we - =0 ( 17.24)
now choose the vertical position of the x-axis so that dx
i.e. there is no elongation of the x-axis. However, even in the situation where the
(17.18) normal force N # 0, we observe from (17.22) and (17.23) that, whereas the bending
moment M is controlled by the term d 2 w/ dx 2 , the normal force is determined by
the quantity du 0 / dx. It is therefore concluded that the phenomena of bending and
If E is constant within the cross-section, this requirement implies that elongation of the beam are uncoupled phenomena, i.e. they can be treated separately.

L zdA = 0
In turn, this means that the FE formulation in the present chapter can be combined
directly with the FE formulation of the elastic bar treated in Chapter 9 if elongation
of the beam occurs. It is emphasized that this advantageous uncoupling is a result
i.e. the x-axis should be placed at the centroid of the cross-section. Equation ( 17.18) of our choice of the vertical position of the x-axis as given by ( 17.18 ).
can therefore be interpreted so that the x-axis is located along the centroid of the
cross-section weighted with the material parameter E. Using ( 17.18), ( 17.17) becomes

M = - d2w
dx2
I A
Ez2 dA ( 17.19) 17.1.5 Differential equations for Bernoulli's beam theory

In the present situation where the normal force N is zero, ( 17.24) holds. Thus ( 17.10)
If E is constant over the cross-section, ( 17.19} reduces to the well-known formula
and ( 17.16) reduce to

M = -EI -d w
2
· I=
d X 2'
I A
z 2 dA (17.20)
( 17.25)

where I is the moment of inertia about the y-axis. The term EI is called the bending
stiffness and when E varies over the cross-section we may for convenience introduce i.e. the axial strain and stress are zero along the x-axis. The neutral axis is in general
the notation defined as that axis along which no axial strain occurs. With the location of the
x-axis defined by ( 17.18) and with the normal force N = 0, the x-axis becomes the

E*I* = L Ez 2 dA (17.21)
neutral axis.
It is of interest to evaluate the term d 2 w/ dx 2 present in (17.22) and (17.25).
From mathematical textbooks on differential geometry we find that the curvature "
where E* I* is again the bending stiffness. This notation means that (17.19) may be of the curve w(x) is defined by
written in the compact form d 2 wj dx 2
K = - - - -- -:--:-:-
(1 + (dw/ dx) 2 ] 312
(17.22) As the slope dw/ dx is assumed to be small, we obtain approximately that
d2 w
i.e. it is possible to express the moment M in terms of the deflection w. K = -- (17.26)
dx2
The normal force N is given by (17.3) which with (17.16) and (17.18} becomes

N= -du
dx
0
I
A
EdA (17.23)
i.e. we may write (17.22) as
M=-E*l*K (17.27)
We observe from (17.22) that it is possible to express the bending moment M
In the present situation we have assumed that N = 0 ( cf. ( 17.4 )). Equation ( 17.23) then in terms of the deflection w. However, the shear stress CJxz cannot be determined using
318 Introduction to the finite element method FE form ulation of beams 319

the constitutive relation since the shear strain Yxz is assumed to be zero. Consequently, The first term is integrated by parts and this results in
the shear force V cannot be expressed by kinematic quantities. Our present objective
is to obtain a differential equation for the beam behaviour expressed in terms of the b-
dv dM [ dM]b fb
- dx = v - +
deflection w. For this purpose we recognize the equilibrium conditions given by ( 17.5)
f a dx dx dx
vqdx
a a
and ( 17.6). As only the moment M can be related to the deflection w, we eliminate Use of(17.6) yields
the shear force V from ( 17.5) and (17.6) to obtain
dv dM dx = [vV]~ + fb vq dx
f bdx
a dx a
(17.28)
The left-hand side of this expression may even be integrated by parts once more to
obtain
It is emphasized that this equilibrium condition holds irrespective of the constitutive
and kinematic assumptions. Use of ( 17.22) in ( 17.28) yields the differential equation
sought: f -
2
- M Jb -
bd2v M dx = [dv [vV]~ -
f bvqdx (17.30)
a dx dx a a

( 17.29) This weak form of the equilibrium equation holds irrespective of constitutive and
kinematic assumptions. As the moment M and the shear force V emerge in the
boundary terms of this weak form, they are the natural boundary conditions.
It appears that when the deflection w has been determined from this differential
equation, all quantities of interest can be derived.
Although Bernoulli's beam theory presented above involves quite a number of
assumptions, it provides close predictions for a variety of problems of practical 17.3 FE formulation
importance. The most critical point in these approximations is the contradiction
between t~ eXiStence ofa shear stress <Jxz - necessar-y to maintain equilibri_um - and From the weak form of the equilibrium condition ( 17.30) we shall now derive the
~~...___._ ... - . - . ~
-·- . .... l

~~~.n Yxz· For long slender beams where the ratio L / h > 5- 10, where corresponding FE formulation. As the deflection w is the unknown function, we can
Lis the beam lengih "and h the beam height, the real shear strain Yxz is generally write the approximation for w quite generally as
small and Bernoulli's theory is usually satisfactory. For higher beams, more refined
beam theories can be used which include the effect of a non-zero shear strain Yxz • w=Na (1 7.31)
for instance Timoshenko beam theory (Przemieniecki, 1968; Timoshenko and Gere,
1972) or Mindlin - Reissner beam theory (Bathe, 1982; Hughes, 1987). For very high where
beams, one has to resort to the general plane stress elasticity theory described in
Chapter 16.
u2
N.]; a= (17.32)

17.2 Weak form of equilibrium equation and n is the number of unknowns for the entire beam. Note that n now denotes the
number of unknowns and not, as in previous chapters, the number of nodal points.
As in the previous chapter, it is advantageous to obtain the weak form of the At this point we shall leave the meaning of the shape functions N; and the nodal
equilibrium condition before the constitutive assumption is introduced. The pertinent values u; unspecified, and merely observe that whereas N; depends on x , i.e. N, = N,(x),
differential equation is given by (17.28), and multiplying by an arbitrary weight the u, components are parameters. From ( 17.31 ) it follows that
function v(x) and integrating over the region (see Figure 17.1) we obtain

fa
bv -
d2M
dx
- dx +
2
fb vq dx = 0
a
[8 X
where (17.33)
320 Introduction to the finite element method FE formulation of beams 321

i.e. the following matrices are defined:

(17.34)

Adopting the Galerkin method, we use the following expression for the arbitrary
weight function v: (17.41)

v = Nc ( 17.35)
With the weight function v being arbitrary, the parameters given by care also arbitrary.
As v = vT, (17.35) can be rewritten as
( 17.36) where K is the stiffness matrix, fb the boundary vector and f1 the load vector. With
( 17.41 ), ( 17.40) may be written in the standard form
which leads to
.... I Ka = fb + f1 I (17.42)
dv TdNT
- =c - - ; ( 17.37)
dx dx As usual, we define the force vector f by

Inserting (17.36) and ( 17.37) into ( 17.30) and noting that the c-matrix is independent (17.43)
of position, we get
i.e.

Ka = f (17.44)

As c is arbitrary, it is concluded that The bqundary conditions for this system of equations are described either by static
conditions, i.e. the moment M and shear force V (the natural boundary conditions),

f b BT Mdx = [dNT M]b - [NT V]~ - fb NT q dx ( 17.38 >.

or by kinematic conditions, i.e. the deflection wand slope dw j dx (the essential boundary
conditions). We therefore have
a dx a a

The similarity with the corresponding forms for two- and three-dimensional elasticity w; dw j d x kinematic boundary conditions
may be noted. We recall that ( 17.38) is an expression for the equilibrium conditi.:>ns M ;V static boundary conditions
for the entire beam and that it holds irrespective of the constitutive model adopted
and irrespective of the kinematic assumptions for the beam response. Examples of such boundary conditions are shown in Figure 17.5 and we observe
At this point it is indeed very easy to derive the FE formulation. From ( 17.22) that static and kinematic boundary conditions may be prescribed at the same
and ( 17.33) we get boundary.

M = - E*/*Ba I ( 17.39)

and insertion into ( 17.38) results in

(17.40)
J7'
w=o w= O W= 0 M=O
M=O M::O dw =O v=0
dx
which is the FE formulation sought. To write this result in a more compact fashion, Figure 17.5 Examples of boundary conditions
 FE formulation of beams 323
322 Introduction to the hnite element method

Let us now consider the FE formulation of one element. With the usual notation, 17.3.1 Completeness and compatibility requirements
the approximation of the deflection w for one element is written as
Up until now we have not specified how the approximati'on ( 17.31) or ( 17.45) was
(17.45) made, and the meaning of the shape function components Ni and the nodal value
w = N•a• components ui have been left undefined. In order to be more specific, we shall
investigate the convergence requirements that are fulfilled if the completeness and
where compatibility requirements are met.
uj For infinitely small beam elements, the deflection within each element will, in
the limit, be given by an arbitrary rigid-body motion superposed by an arbitrary
ui (17.46) constant curvature. We note that a constant curvature is the simplest deflection state
N• = [N~ Ni ··· N~J; a• =
that creates strains (cf. (17.25)). We are then led to the following completeness
u•n, requirements:

where n. is the number of unknowns for one element. Again we deviate from the
notation in the previous chapters and emphasize that n. is the number of unknowns
and not the number of nodal points. We define Completeness
• The approximation for the deflection w must be able to represent an
(17.47) arbitrary rigid-body motion.
• The approximation for the deflection w must be able to represent an
arbitrary constant curvature.
It is obvious that

K•a • = r• (17.48)

An arbitrary rigid-body motion consists of a translation in the z-direction as well as

where
a rotation in the xz-plane ( cf. Figure 17.1 ). That is, the approximation for the deflection
(17.49) w must at least include the expression ct 1 + ct 2 x where ct 1 and ct 2 are arbitrary
parameters. The curvature K is given by (17.26) and we therefore conclude that any
approximation for the deflection w must take the form
and

K• = f BeT E* J*B• dx (17.51)

L,

f~ = [N•TVk - [ d~;T M 1. (17.50) To derive the compatibility criterion, we shall take advantage of the concept of
C"-continuity discussed in Chapter 9 (cf. page 202). In all our previous applications

ff = f L,
N•Tq dx
of the FE method, the B-matrix in the stiffness matrix K contained only first derivatives
of the shape functions. Therefore, C 0 -continuity was required and, in turn, this implied
the compatibility condition that the approximation of the variable must vary in a
where La is the region of the element and the notation defined by (9.79) has been continuous manner over the element boundaries. According to (17.41) and (17.33),
used in the expression for f~. We recall that the boundary conditions for an element the stiffness matrix now involves second derivatives of the shape functions. We
are unknown except when an element boundary coincides with the boundary of the immediately conclude that C 1 -continuity is required for the shape functions and thus
beam. also for the deflection w. This compatibility requirement can be stated as follows:
324 Introduction to the finite element method FE formulation of beams 325

In order to determine the parameters ct 1 , ct 2 , ct 3 and ct4 in (17.53), the C-matrix

Compatibility method of Chapter 7 is applied ; cf. for instance (7.45 )-(7.52 ). Therefore, ( 17.53 ) is
written as
• The approximation for the deflection w must vary
(17.52)
continuously and with continuous slopes over the element w=N~ (17.54)
boundaries.
where

Previously, we claimed that non-conforming elements are acceptable (cf. p. 93 ).

However, we note that Bernoulli beam elements must always be compatible since a N = [1 ( 17.55 )
possible discontinuity does not disappear for infinitely small elements.

17.3.2 Simplest possible beam element We next express the conditions for w at the nodal points, i.e.
-v·~ r ''lk
l2j11-Y--/JJ 1P O-
We shall now establish the simplest possible compatible beam element. Based on
( 17.56 )
(17.51), we choose the approximation for the deflection was
(17.53)
Use of these conditions in ( 17.54) results in
where a 1 , ct 2 , a 3 and a 4 are certain parameters. This means that the completeness
criterion is fulfilled. To determine the four parameters in ( 17.53 ), the beam element a• = c~ (1 7.57)
must possess four unknowns and these unknowns are taken as the quantities u 1 , u 2 ,
u 3 and u4 shown in Figure 17.6. Here, u 1 and u 3 are the nodal displacements in the where
z-direction, whereas u 2 and u4 are the slopes at the nodal points measured as positive u1 1 0 0 0
in the counter-clockwise direction. It appears that two unknowns are related to each
nodal point: one for the deflection wand another for the slope dw j dx. u2 0 0 0
a• = C= (17.58)
As values for w and dwj dx are connected to each nodal point and as such a u3 t( L L2 L3
nodal point is common for two neighbouring elements, not only the deflection w but ·1
u4 0 2L 3L2
also its derivative dw j dx will vary in a continuous manner over the element
boundaries, i.e. the simple beam element of Figure 17.6 fulfils the compatibility
requirement given by ( 17.52). From ( 17.57) it follows that
1 0 0 0
0 1 0 0
~=C- 1 a• where c - 1 = (17.59)
2 2
- 3/ L -2/ L 3/ L -1 / L
2/ L 3 1/ L2 -2/ L 3 1/ L 2

and use of this result in ( 17.54) gives

w = N•a• ( 17.60)
L
·I where
Figure 17.6 Simplest possible beam element N• = :NC- 1 = [N1 Ni N~ N~] (17.61 )
326 Introduction to the finite element method FE formulation of beams 327

1~ ~1
f---..- X t---X
L L

~
dNe __ 2 "1
dx dNe
__
4 =1 Figure 17.9 Second derivatives of element shape functio ns
dx

Figure 17.7 Element shape functions

u, u3 us

dNr dN~
u2 rh u4 rh
0
u5rh
dx dx
l/2 L/ 2
·I
L

d Ne
_ _2_ 1~
t __ dN! ~1
~--_j
dx dx Figure 17.10 Three-node beam element

Figure 17.8 First derivatives of element shape functions

characteristics. The derivatives dNi f dx and dN~ jdx, however, possess the feat ures
From this expression we derive of the element shape functions dealt with in the previous chapters; for instance, we
have that (dNi f dx)x;o = 1 and (dNi / dx )x;L = 0. The properties of the element

x 3-2 -x)
N3= -
U
2

L
(
shape functions shown in Figure 17.7 and 17.8 are clearly necessary so that the
) conditions ( 17.56) can be fulfilled.
(17.62) We may consider beam elements of higher order, and Figure 17.10 shows an
x (xL- 1)
N~ =L
2
example with three nodal points and a total of six degrees of freedom. The
approximation for the displacement w is given by
(1 7.64)
and these element shape functions are illustrated in Figure 17.7. From ( 17.60) it
follows that Irrespective of the beam element considered, it follows directly that the global
shape functions can be established from the element shape functions in the same
(17.63) manner as discussed in Chapter 7. For more details about beam elements, we refer
to Przemieniecki ( 1968) and Bathe ( 1982), which even provide the FE formulation
As with previous FE formulations, it appears that the deflection w within the element of the Timoshenko beam theory discussed previously.
is composed of contributions from the element shape functions N~, Ni, N'3 and N~ Let us return to the topic discussed above, namely that the B-matrix now contains
weighted with the nodal quantities Uu u 2 , u3 and u4 . The first and second derivatives second derivatives of the shape functions, whereas in the previous chapters all
of the element shape functions are shown in Figures 17.8 and 17.9, respectively. From B-matrices only involved first derivatives of the shape functions. The reason is that
(17.39) and Figure 17.9 it follows that the simple beam element is able to model an the differential equation for the !?.~am (cf. (17.29)) i§ of fourth order, whereas all the
arbitrary linear moment variation within the element. prewu; 'pi:lysical problems were governed. by differential equations of second order.
It appears that, whereas the element shape functions N~ and N'3 related to the In accordance with this, the weak form of the equilibrium conditions for the beam
displacement degrees of freedom possess the traditional properties, the element shape ( cf. ( 17.30)) was obtained by invoking integration by parts twice, which implied that
functions Ni and N~ related to the rotational degrees of freedom possess new the order of differentiation of the weight function became two. As a consequence, the
328 Introduction to the finite element method FE formulation of beams 329
1
approximation for the deflection w was required to be C -continuous in contrast to 17.3.4 Evaluation of element boundary vector
the previous chapters where only C 0 -continuity was demanded. The C 1 -continuity
means that the deflection w varies continuously and with continuous first derivatives- Consider the element boundary vector f~ given by ( 17.50). Evaluating this vector for
i.e. slopes - over the element boundaries. Approximations of functions which fulfil the simple beam element shown in Figure 17.6, we obtain
not only the function value itself at some specific points but also its slopes at these
points are called Hermite interpolations and a simple example was given by ( 17.62)
and (17.63). f~ = (N•T V]~ - [dN•T M]L
dx 0

i.e.

17 .3.3 Evaluation of element stiffness matrix (N~ V)x=L - (N~ V)x=o - 1

(dN M) + (dN1 M)
dx x=L dx x=O
From ( 17.50) we obtain for the simple beam element given by Figure 17.6
(Ni V)x =L - (Ni V)x=o - (dNi M) + (dNi M)
B~ B~ B~Bi B~Bj B~B~
dx x=L dx x=O
f~ =

K· =I: E*I*
BiB~

Bj B~
BiBi
BjBi
BiB3 Bi B~
BjBj BjB~
dx (17.65)
(Nj V)x =L- (Nj V)x=o- (dNj M)
dx x=L
+ (dNj M)
dx x=O

B~B~ B~Bi B~Bj B~B~ (N~ V)xzL- (N~ V)x =o- (dN~ M) + (dN~ M)
dx x=L dx x;O
We now assume that the bending stiffness E* I* does not vary along the beam axis;
then, for example, the component Kj 2 becomes Considering the properties of the element shape functions given by Figures 17.7 and
17.8, it follows that
L d 2 N·3d 2 N·2
K•32 -- B•3 E* I* B•2 dx = E* I* JL B•3B•2 dx = E* I* IL -dx2-- dx2- dx
J0 0 0 - Vx=O
M x=O
From ( 17.62) it follows that f~ = ( 17.67)
Vx =L

Kj 2 = E* I*
I0
L (
- 62 - -12x)(
L L 3 - -
4
L
+ -6x) dx = - -
U
6E*-
L2
I* -Mx=L

The positive directions of V and M are given in Figure 17.2. Therefore, when all the
Evaluating all the components of ( 17.65) in this manner we arrive at components of the element boundary vector fb are positive, these components take
the form shown in Figure 17.11. It appears that the directions shown in Figure 17.11
12E* I* 6E*I* 12£* I* 6E*I* are in accordance with the directions of the kinematic quantities shown in Figure 17.6.
L3 L2 LJ L2
6E*I* 4E*I* 6E* I* 2E*I*
L2 - -L2-
L L f~, f~3
K•= ( 17.66)
12E* I*
LJ
6E* I*
- --
L2
12£* I*
LJ
6E*I*
- -L2- (~211~ f~4 rh
6E* I* 2E*I* 6E* I*
- -L2-
4E*I* I· ·I
L2 L L
Figure 17.1 I Illustration of element boundary vector fi,
330 Introduction to the finite element method FE form ula tion of beams 331
17 .3.5 Beam element with no distributed loading ( 17.68) with K• given by ( 17.66) we obtain

Consider the simple beam element with no distributed loading, i.e. q = 0. This implies 12£* I*
that ff = 0 and that ( 17.48) reduces to L3
K•a• = f~ (17.68) 6E* I*
where ~ is given by ( 17.67) and the positive directions of the ~-components are L2
shown in Figure 17.11. Assume now that the bending stiffness E* I* does not vary f~ = ( 17.70)
12E* I*
along the beam axis, i.e. K• is given by (17.66). From (17.68) we may obtain some
interesting results. Assume that L3
6E*I*

0
u
a• = (17.69)
0 These <:omponents are illustrated in Figure 17.12(b ). Choosing next u2 = 1 and
u 1 = u 3 = u4 = 0, we derive the results shown in Figure 17.12( c) and (d). Proceeding
0
in this way, we arrive at all the results shown in Figure 17.12.
With the a•-components given by Figure 17.6, the nodal values of ( 17.69) correspond In comparison with standard results in textbooks on beam theory, it appears
to the deflection of the beam element as shown in Figure 17.12(a). Using (17.69) in that the results of Figure 17.12 are precisely those derived directly from beam theory.
The reason why our FE solution provides the exact solutions and not just approximate
a) ones is easily explained. With the bending stiffness E * I* independent of x and with
no distributed load, i.e. q = 0, the differential equation ( 17.29) becomes
d4 w
E*I* - - = 0
> dx 4
with the solution
c) d) 6elx 6Exix
~ ---cr- w· = C 1 + C 2 x + C3 x 2 + C4 x 3

> 4rr·c'=====t) 2E:r· where C 1 ••. C 4 are arbitrary constants. However, this expression for the deflection
is exactly the same as the one our F E solution was based on ( cf. ( 17.53)). We conclude
that if q = 0 and if E* I* is independent of x, the simple beam element provides exact
results and not just approximations. In general, however, use of this simple element
e) f) 12E'Ix 12E'Ix will give results which are approximations to the beam theory.
-Ll ~

~----- - ) > _6E;Ix(t= ====f ) _6LE;r• 17.3.6 Evaluation of element load vector - uniform load
u 3= 1 L
\ Assume that the simple beam element of Figure 17.6 is loaded by a uniform load q;
sE•r• ser• I
g) h) I cf. Figure 17.13. The element load vector ff is to be calculated. From ( 17.50) we obtain
T -L2
N1
> _2~_xlx ( t=====t) 4E:r• ff = LL N•T q dx = q f: ~! dx
Figure 17. 12 Deformations and corresponding forces and moments N~
332 Introduction to the finite element method FE formulation of beams 333
q"'\
tttttttt M = Pb

L
I· ·I ,__.__X
x=a

Figure 17.13 Uniform load q and equivalent nodal forces and nodal moments x=a x= a+b .j

and use of ( 17.62) implies that Figure 17.15 Equivalence between moment M and force couple

!qL
-f2qLz
ff = (17.71) M
!qL n
-/2qLz
These equivalent nodal forces and nodal moments are illustrated in Figure 17.13.
X=O

Figure 17.16 Moment M and equivalent nodal forces and nodal moments for simple beam
element
17.3.7 Evaluation of element load vector- concentrated force
and moment i.e.

Let us evaluate the element load vector ff for an arbitrary type of beam element (17.72)
when the load q takes the form of a concentrated force P; see Figure 17.14. The
dimension of P is [N] and P is measured as positive in the z-direction. As before, If the simple beam element is considered, the components of ff are as illustrated in
we use Dirac's delta function and write q as Figure 17.14. In general, the force P gives rise to both nodal forces and nodal moments.

q = P<5(x- a); i.e. IL q dx = IL P()(x - a) dx = P

As easily shown by means of ( 17.72) and Figure 17.7, the - ...... ...- where the
only situation
.....--...._,_.___
19!£~~!CE.S !lOd<!l forces but zero nod~l QlQrnent_s is when the force is located at
one of the end points of .the simple beam element. •
~- .-
~ ~

Another situation which often occurs in practice is the application of a

This implies that ff given by ( 17.50) becomes moment M . For an arbitrary beam element, let us evaluate the element load vector
ff for such a loading. The moment M with the dimension [N m] is measured as
f)= LN•Tqdx = IL N•Tp(j(x - .a)dx = P IL N•T(j(x- a)dx positive in the counter-clockwise direction (see Figure 17.15). It is obvious that the
I
0 0 0 effect of this moment is equivalent to the force couple shown in Figure 17.15. Here
the distance b between the two opposite directed forces is taken to be infinitely small.
Referring to ( 17.72 ), the element load vector f) for the force couple of
p PN~(x=a)
Figure 17.15 is

I (17.73)
As b denotes an infinitesimal distance, a Taylor expansion of N•T about the point
1----X
x =a gives
x=a
.I
Figure 17.14 Concentrated force P and equivalent nodal forces and nodal moments
N~~a+b ~ N~~a + (dN•T) b
dx x;a
334 Introduction to the finite element method

Use of this expression in ( 17.73) yields

rIs __~-----
fj = M(dN•T)
dx x=a
(17.74) 1 FE formulation of plates I
where M = Pb (cf. Figure 17.15). r ......._
. !f the simple· beam element is considered, the components of fj are as illustrated \
tn Figure 17.16. It appears that the moment M, in general, gives rise to both nodal
forces _and ?odal moments. As easily shown by means of ( 17.74) and Figure 17.8, the ·
onl~u-~_!!o~ whe~e the ~oment M creates nodal moments b~J. zero no_®Lforces is
~--~omen~ ~ applied at one <;>f the end po(nts of the simple beam element.
' '-

Following the treatment of the engineering theory of beams it is natural to consider

the engineering theory of plates. Again we will discover that the engineering theory
of plates is not a theory that fulfils all the field equations. This means that the theory
of plates is an engineering approximation which reduces the original three-dimensional
problem to a simpler two-dimensional problem. For many applications, however,
this engineering theory provides realistic solutions, especially if the plate is thin.
Generally speaking, a plate is a structure with a thickness t that is small compared
with all other dimensions of the plate; see Figure 18.1. Moreover, a plate is loaded
by forces normal to the plane of the plate. To be more specific, we introduce a
coordinate system and assume that the configuration of the plate is symmetric about
the xy-plane, as shown in Figure 18.1. This means that the xy-plane is located in the
mid-plane of the plate, and even though the plate thickness tin the z-direction may
vary, it varies symmetrically about the xy-plane. The plate is loaded by a transverse
loading q measured as positive in the z-direction. This loading q is force per unit
area, i.e. q has the dimension [N/ m 2 ]. The deflection w of the plate is measured as
positive in the z-direction.

18.1 Plate theory

18.1 .1 Equilibrium conditions
For sections normal to the x- and y-axes, the stress components (}"'"'' (}"'"' (Jxz and
exist. These stress components give rise to the following forces and
(Jyx• (JYY' (Jyz

moments:

t/2
Vxz =
f
- t/ 2
t/ 2
(Jxz dz
( 18.1)
V,, =
f
- t/ 2
(}1 z dz

335
334 Introduction to the finite element method

Use of this expression in (17. 73) yields 18

ff = M(dNeT)
dx x=a

where M = Pb (cf. Figure 17.15).

(17.74)

If the simple beam element is considered, the components of ff are as illustrated

[ FE formulation of plates
I
in Figure 17.16. It appears that the moment M , in general, gives rise to both nodal
forces and nodal moments. As easily shown by means of(17.74) and Figure 17.8, the
only situation where the moment M creates nodal moments but zero nodal forces is
when the moment is applied at one of the end points of the simple beam element.

Following the treatment of the engineering theory of beams it is natural to consider

the engineering theory of plates. Again we will discover that the engineering theory
of plates is not a theory that fulfils all the field equations. This means that the theory
of plates is an engineering approximation which reduces the original three-dimensional
problem to a simpler two-dimensional problem. For many applications, however,
this engineering theory provides realistic solutions, especially if the plate is thin.
Generally speaking, a plate is a structure with a thickness t that is small compared
with all other dimensions of the plate; see Figure 18.1. Moreover, a plate is loaded
by forces normal to the plane of the plate. To be more specific, we introduce a
coordinate system and assume that the configuration of the plate is symmetric about
the xy-plane, as shown in Figure 18.1. This means that the xy-plane is located in the
mid-plane of the plate, and even though the plate thickness t in the z-direction may
vary, it varies symmetrically about the xy-plane. The plate is loaded by a transverse
loading q measured as positive in the z-direction. This loading q is force per unit
area, i.e. q has the dimension [N/m 2 ]. The deflection w of the plate is measured as
positive in the z-direction.

18.1 Plate theory

18.1.1 Equilibrium conditions
For sections normal to the x- and y-axes, the stress components (Jxx• (Jxy• (Jxz and
exist. These stress components give rise to the following forces and
(Jyx• (JYY' (JY•

moments:
t/ 1
Vxz =
f
- t/2
t/ 2
(Jxz dz
( 18.1)
Yy. =
f
- t/1
(JY• dz

335

I
336 Introduction to the finite element method FE formulation of plates 337
Z,W

>

Figure 18.3 Illustration of M yy• Myx and Yy,

Figure 18.1 Configuration and loading of plate

X
)-,
Figure 18.4 Illustration of horizontal forces N XX> Nyy and N xy

components ayx• aYY and a yz acting on this plane create the twisting moment Myx
per unit length, the bending moment M yy per unit length and the vertical shear force
Figure 18.2 Illustration of Mxx• Mxy and Vxz v;,z per unit length. The positive directions of these quantities are shown in Figure 18.3.
In addition to these moments and vertical shear forces, we note that the stress
components a xx• a YY and a xy also result in the following horizontal forces:
and
t/ 2

Mxx =
t/2

f
- r/2
UJxx dz
Nxx =
f - t/2
t/2
<1xx dz

Myy =
t/ 2

f
- t/2
zaYY dz ( 18.2)
N yy =
f - t/ 2
aYY dz

t/2
( 18.3)

Nxy = Nxy =
f - t/2
(Jxy dz

These forces act in the xy-plane as shown in Figure 18.4. It appears that Nxx is the
To illustrate the quantities defined by (18.1) and (18.2), consider first a plane cut norma/force per unit length in the x-direction, NYY is the norma/force per unit length
in the plate and normal to the x-axis; see Figure 18.2. The stress components axx• in the y-direction and N xy is the horizontal shear force per unit length.
axy and <Jxz acting on this plane create the quantities Mxx • Mxy and Vxz defined by Let us now establish the equilibrium conditions for the plate. The plate is assumed
(18.1) and (18.2) and shown in Figure 18.2.1t appears that Mxx is the bending moment to be loaded by transverse forces only. Therefore, as no resulting forces (or restraints)
per unit length, M xy is the twisting moment per unit length and Vxz is the vertical act in the xy-plane, horizontal equilibrium requires that
shear force per unit length. The positive directions for Mxx• Mxy and V.z are also
shown in Figure 18.2, and to illustrate the moments in a convenient manner, the Nxx = Ny y = N xy = 0 (18.4)
standard conventional double-headed arrows has been adopted. Consider now an infinitesimally small part ofthe plate, Figure 18.5. As no resulting
Considering next a plane cut in the plate and normal to the y-axis, the stress forces act in the xy-plane {cf. (18.4)), all forces and moments acting on this small

I
338 Introduction to the finite element method
FE formulation of plates 339
dy As dy is an infinitely small quantity we get

( 18.6)

aVyz Considering the moment equilibrium about one of the sides parallel to the y-axis,
z Vyz+ - - dy we derive in a similar manner from Figure 18.5 that
ay

( 18.7)

18.1.2 Kinematic relations

The plate is assumed to deform in accordance with Bernoulli's assumption for beam
behaviour, i.e. that plane sections normal to the mid-plane remain plane and normal
to the mid-plane during deformation. In complete analogy with Figure 17.4 and
(17.7), we obtain the following displacements in the x-, y- and z-plane directions:
Figure 18.5 Vertical shear forces and moments acting on an infinitesimally small part of the
plate
u = u - z-
0 OW
.x OX
part of the plate appear from Figure 18.5. Vertical equilibrium requires that
0 ow
u = v -z -
~. dx + ( v,. + a~. dx) dy + ( ~. + a:;· dy) dx - v,. dy = 0
( 18.8)
q dx dy - >' oy
i.e.
where u 0 and v 0 are the displacements of the mid-plane in the x- and y-directions,
av,. a~.
- + - + q=O respectively. It follows that
ax ay (18.5)
u0 = u0 (x , y); v0 = v0 (x, y) and w = w(x, y) (18.9)
Considering next the moment equilibrium about the right side of the small part where we have introduced the assumption that the deflection w is independent of z,
of the plate, which is parallel to the x-axis, we obtain i.e. w = w(x, y). From (18.8) and (18.9), the strains as given by (12.37) and (12.40)
become
- q dx dy t dy + ~z dx dy- ( vxz + a~. d x ) dy t dy + vxzdy t dy ou0 - z -
8 w 2
=-
e
xx ox ox-2
+ M >'>' dx - ( M .xy + _oMx_>' dx ) dy - ( M oM dy ) dx + M
+~ dy = 0
~ " ~ Q
ov 0 o2 w
e = - - z -- (18.10)
i.e. >'>' oy oy 2

y
ou0 ov 0 o2 w
= - + - -2z--
.xy oy ax •axay

l
340 Introduction to the finite element method FE formulation of plates 341
and but we shall for simplicity consider the constitutive relation in the form given by
ezz = Yxz = Yyz = 0 (18.11) ( 18.12).

We note that the kinematic assumptions imply that the shear strains Yxz and Yyz are
zero. 18.1.4 Further derivations

The strains (18.10) may be written in the form

18.1.3 Constitutive relation ( 18.16)

Hooke's law is assumed to be applicable. As in the discussion of beam behaviour where

given in relation to (17.12)-(17.14), it is not possible to obtain a correspondence au 0 a2w
between the non-zero shear stresses ax. and ayz necessary to maintain equilibrium
(cf. (18.5)) and the zero shear strains Yxz and Yyz· However, we assume the plate to ax ax 2
be thin, i.e. the largest stresses will be a""' a yy and a xy that are illustrated in Figure 18.6.
=
av 0 o2w (18.17)
llo K=
This observation suggests that the assumption of plane stress is most applicable, ay ay2
i.e.
au 0
av 0 o2 w
-+- 2-
I (J = Dt I (18.12) ay ax axay

where and K is denoted the curvature matrix; cf. the discussion of (17.26). We emphasize
that e 0 expresses the straining of the mid-plane and should not be confused with the
initial strains given by 8 0 . Using (18.16) and (18.12) we get
(J = [:::]; t = [:::] (18.13)
(18.18)
(Jx y Yxy
We define the matrix M by
For isotropic elasticity, the plane stress constitutive matrix Dis given by (13.33), i.e.

D = - E2
1
[~ ; ~ ] (18.14) M=
[M""]
Myy ( 18.19)
- v 0 0 1(1 - v) Mxy

In the following, however, the D-matrix may take any form. We also observe that it
Then the expression for the moments given by ( 18.2) can be written in the compact form
is even possible to consider initial strains, llo for instance, in terms of thermal strains.

f
In this case (18.12} is replaced by t/ 2
M = az dz ( 18.20)
a= DB- DB0 (18.15) -t/2

Inserting (18.18) into (18.20} and recalling that neither 8° nor K depends on the
coordinate z yield
t/2 ft/2
M = Dt 0 f -t/2 z dz - DK z 2 dz (18.21)
~ -t/2

where it was assumed that D is independent of z, i.e.

Figure 18.6 Illustration of stresses uxx• u yy and u xy D = D(x,y) ( 18.22)

i
342 Introduction to the finite element method FE formulation of plates 343

We have The equilibrium conditions were given by ( 18.5)-( 18.7). As only the moments
can be expressed in terms of kinematic quantities, we eliminate the shear forces Vxz
t/2 J''2z t3 and V,z from (18.5)-(18.7). Therefore, we differentiate (18.6) with respect to y and
I - r/2
z dz = 0;
- r/2
2
dz =-
12 (18.7) with respect to x; we add the results and use (18.5) to obtain
That is, ir("espective of the value of &0 -strains, (18.21} reduces to

I M ~ - ll• I whece EiEJ ( 18.23)

( 18.27)

This equilibrium condition holds irrespective of the constitutive assumption. Let us

* defined by
With a defined by ( 18.13 }, the horizontal forces given by ( 18.3) can be expressed as now introduce the matrix differential operator V

XX 1/ 2
N ]
N>'>'
[ Nxy
=I -t/ 2
adz ox 2
()2
*
V= (18.28)
Use of (18.18) yields oy2
02
Nxx]
Nn = De t 0
( 18.24)
2-
oxoy
[ Nxy It appears that the equilibrium condition (18.27) may then be written as
In the present case, no resulting horizontal forces act in the mid-plane ( cf. ( 18.4)}
( 18.29)
and we conclude that
( 18.25) Moreover, we may rewrite ( 18.17) as
In accordance with expectations, this shows that there is no straining of the mid-plane. *
K=Vw ( 18.30)
However, even in the situation where the horizontal forces are different from zero,
we observe from (18.23) and (18.24) that, whereas the moments Mare controlled by i.e. (18.23) takes the form
the curvature matrix K, the horizontal forces are determined by the in-plane strains
0
& . This implies that the phenomena of bending and straining of the mid-plane are
uncoupled phenomena, that can be treated separately if ( 18.22) holds. If the mid-plane
M= - DVw I (18.31)

is deformed, i.e. & 0 i= 0, we have so-called membrane action. It is concluded that, if Introducing this expression into ( 18.29) yields the following differential equation
membrane action is present, the FE formulation of the present chapter can be for plate theory :
combined directly with the FE formulation of plane stress elasticity treated in Chapter
16. This uncoupled response is similar to that observed for beam theory, and it implies
that shell structures in which membrane action is of prominent importance may be
l VTf>Vw = q ( 18.32)

analyzed by a shell element which comprises a combined plate and plane stress element.
When the deflection w has been determined from this differential equation, all
quantities of interest can be derived. Assume now that the thickness t .is con~~ant
18.1.5 Differential equations for plate theory and that isotropic elasticity is considered, i.e. D is given by (18.14). If, m add1t10n,
D does not depend on x and y, ( 18.32) takes the form
In the present situation where no resulting horizontal forces act in the mid-plane,
(18.25) holds, i.e. (18.16} and (18.18) reduce to
( 18.33 )
& = -ZK and a = - zDK ( 18.26)

I
344 Introduction to the finite element method FE formulation of plates 345

This differential equation is ~ailed the biharmonic equation and it was derived in 1811 by have
Lagrange, even though hts plate theory was not entirely satisfactory. The first
convincing plate theory was established by Kirchhoff(1850) and what was presented
( 18.34)
above is essentially due to him. Th_is plate theory is therefore also termed Kirchhoff
plate t_heory and we may ~efer, for ms~ance, to the textbooks by Boresi et al. (1978)
and T1moshenko and Womowsky-Kneger (1959) for more details. The most critical
point in the Kirchhoff theory is the contradiction between the existence of the shear
stresses axz and ayz - necessary to maintain equilibrium - and the zero shear strains and
'Yxz and /'yz· In an~logy with the Bernoulli beam theory, the Kirchhoff plate theory (18.35)
w~rks well for thm ~lates where the real shear strains 'Yxz and 'Yyz are small. For
th1ck:r plates, the_Kirchhoff ~heory becomes questionable, and more refined plate The traction vector t acting on the section defined by n is given by (12.14), and as
theones must be mvoked wh1ch allow for the existence of non-zero shear strains.
Among such refined plate theories, the one proposed by Mindlin ( 1951) and Reissner n. = 0, we get
(1945) is t?e most prominent. This Mindlin- Reissner plate theory is also interesting
from a fimte element point of view since only C 0 -continuity is required, whereas -
as we shall see - the Kirchhoff theory requires C 1 -continuity. Here we shall not treat
tx] [(Jxx nx + (Jxyny] (18.36)
= ayxnx + ayyny
the ~E formulation ofMindlin-Reissner plates, but the interested reader may consult,
for mstance, Hughes (1987) and Cook et al. (1989).
t
[tz = (Jzxnx
ty
+ (Jzyny

Previously, when defining stress components, we considered, for instance, a plane

normal to the x-axis, and the stress components axx• axy and a xz were defined as the
compL '1ents of the traction vector t in the x-, y- and z-directions, respectively; cf.
18.1.6 Moments and shear forces acting on an arbitrary plane Figure 12.2 and expression (12.2). In a similar fashion, for the plane identified by the
unit normal vector n we now define the stress components a"" ' anm and anz as the
Consider a section in the plate normal to the mid-plane and defined by the unit components of the traction vector in the n-, m- and z-directions, respectively. These
normal vector n, which is located in the xy-plane (see Figure 18.7). A unit vector m
stress components are illustrated in Figure 18.7 and we obtain
located in the xy-plane is also defined and taken to be tangential to the section, i.e.
orthogonal to n. Moreover, the vector m is chosen so that n and m after a suitable
rigid-body rotation, follow the same directions as the x- andy-axes, ~espectively. We (inn= DTt
(Jnm = mTt
(Jnz = [0 (} 1]t
z
With nand m defined by ( 18.34) and the traction vector t by ( 18.36) this result becomes

ann= n;axx + n;ayy + 2nxnyaxy

(jnm = nxmxaxx + nymyayy + (nymx + nxmy)axy ( 18.37)
m

We may note that the first two equations of ( 18.37) correspond in fact to the results
n obtained by the well-known Mohr's circles of stress.
As in the definitions (18.1) and (18.2 ), we define the bending moment Mnn per
Figure 18.7 Illustration of stress components u""' u.m and u., in a plane defined by the unit
normal vector n
unit length, twisting moment Mnm per unit length and vertical shear force V,. per

I
346 Introduction to the finite element method
FE formulation of plates 347
unit length acting on the section defined by the unit vector n according to
As n and m are orthogonal, any vector r in the xy-plane may be expressed as
t/2
M •• =
f - t/ 2
t/ 2
za"" dz
r = t:xn +{3m
where t:x and f3 are some parameters. Since n and m are orthogonal unit vectors, it
appears that
Mnm =
f -t/ 2
t/ 2
ZlTnm dz ( 18.38) Rr = r; i.e. (R - l)r = 0
As this relation holds for arbitrary vectors r we conclude that R = I, i.e.
V,z =
f - t/ 2
(1 nz dz
nnT + mmT =I (18.44)

Inserting (18.37) into (18.38) and making use of the definitions (18.1) and (18.2) where I is the unit square matrix. That is (18.43) reduces to
provide
nM"" + mM.m =An
M"" = n;Mxx + n;Myy + 2nxnyMxy Multiplication by the quantity (Vv)T, where v = v(x, y) is an arbitrary function and
Vv denotes the gradient of v, yields
Mnm = nxmxMxx + n1 m1 Myy + (n 1 mx + nxmy)Mx1 (18.39)
(Vv)TnM"" + (Vv)TmMnm = (Vv)T An
V..z = nx V.:z + ny Yyz
Use of (5.8) and (18.42) gives the following relation, which will later be used in
It may be observed that these expressions are completely similar to ( 18.37). Therefore, the weak formulation:
just as it is possible to identify the principle stresses, i.e. sections for which the shear
stress anm = 0, it is possible to determine the principle moments, i.e. sections for which
dv Mnn + dv Mnm = [ ov ov] [Mxxnx + Mxyny] ( 18.45)
dn dm OX oy Mxynx + Myyny
the twisting moment Mnm = 0. We shall not pursue this subject, but refer to standard
textbooks for details.
For future purposes, some useful relations will be derived from (18.39). From
the three-dimensional vectors n and m defined by ( 18.34) and ( 18.35), we define the
two-dimensional orthogonal vectors n and m by
18.1. 7 Discussion of proper boundary conditions
n = [ :: J m = [ :: J Inl = Im I = 1; nTm = 0 ( 18.40)
The boundary conditions for Kirchhoff plate theory turn out to be non-trivial. In
reality, the identification of proper boundary conditions had already bee~ established
These two unit vectors are located in the xy-plane. By carrying out the matrix by Kirchhoff himself in 1850, but even today a profound understan~mg of the~e
multiplications it follows that
boundary conditions is something which most students do not fully achteve. For thts
reason, we shall first try to illuminate the problem and then, in the next section,
( 18.41)
provide the solution.
where From previous results we have arrived at a two-dimensional problem, where
A= [Mxx Mxy] ( 18.42)
everything depends on the coordinates x and y in the mid-plane of the plate. The
region spanned by this mid-plane is denoted by A and its boundary by 2; see
M x1 M 1 y
Figure 18.8. . .
Using ( 18.40)- ( 18.42) the following expression is derived: The assumption for the kinematics of the plate is such that a stratght bne normal
to the mid-plane remains straight during deformation. Along the boundary 2, the
nMnn + mMnm = (nnT + mmT)An (18.43) movement of such a straight line is given by its slope e. normal to the boundary
Let us investigate the square matrix R defined by and its slope em tangential to the boundary ; see Figure 18.9. Therefore, along the
boundary we can identify the following three kinematic quantities: the defl~ctio~ w
R = nnT + mmT and the slopes en and em. lh Kirchhoff plate theory, it is assumed that a stratght line

I
 r
348 Introduction to the finite element method FE f ormulation of p lates 349
y two independent kinematic boundary conditions given by (18.46), we apparently have
m
three independent static boundary conditions ( 18.47).
n
To investigate this problem, consider the differential equation in w given by
(18.32). Without loss of generality we may for simplicity consider the biharmonic
differential equation ( 18.33) valid for isotropic elasticity. It appears that this differential
equation is of fourth order and we may recall that the differential equation for beam
theory is also of fourth order. Assuming for convenience that the boundary It' consists
of two regions, one part where the kinematic boundary conditions are prescribed
Z~--------------------------------x and another part where the static boundary conditions are prescribed, we are led
Figure 18.8 Two-dimensional problem formulation of plate theory to the conclusion that, along each of these parts, the solution of the fourth-order
differential equation requires the specification of two independent boundary
conditions. This requirement is fulfilled for the kinematic boundary conditions ( 18.46),
and it implies that the three static quantities M •• , M.m and V..z must combine in such
a fashion that they appear in the form of two independent terms. Therefore, we must

z ~~
~V-dn
•1 dw
establish a way in which we can identify these two independent static boundary
conditions. We have previously seen that the weak form by itself gives the natural
boundary conditions, so it is tempting to obtain a weak form of the plate problem
WI~;) and in this fashion identify the proper static boundary conditions. This will be outlined
in the next section.
- ---~n Before deriving the weak form, we mention that the reason why we have 'lost '
one static boundary condition is due to the reformulation ofthe equilibrium equations.
Figure 18.9 Kinematic quantities w, 8. and 8m along the boundary Originally, we had the three first-order differential equations ( 18.5)- ( 18.7)
corresponding to three independent static boundary conditions. However, when
eliminating the shear forces we arrived at the second-order differential equation
normal to the mid-plane remains straight and normal to the mid-plane during ( 18.27), which only requires two static boundary conditions.
deformation, i.e. we have e.= dw/dn and em= dw/ dm. That is, the three kinematic We also mention that in higher-order plate theories like the Mindlin- Reissner
quantities that we can identify are w, dw/dn, dw/dm. Suppose now that the deflection theory, the formulation is based on (18.5}-(18.7) and not on (18.27), i.e. we have
w is known along part of the boundary. We may write this as w = w(m); cf. three independent static boundary conditions given by M •• , M.m and Vnz· Moreover,
Figure 18.8. However, if w(m) is known along the boundary, then the slope dw / dm as shear strains are now included, we also have three independent kinematic boundary
is also known. This means that we only have two independent kinematic quantities conditions given by the deflection wand the slopes and e. emwhere it now holds that
that we may prescribe, namely w and dw / dn. These may be termed the kinematic
boundary conditions, i.e.
e."' dw / dn and em"' dw/ dm.

(18.46)
18.2 Weak form of equilibrium equation -
proper static boundary conditions
are the independent kinematic boundary conditions. We shall now establish the weak form of the equilibrium equation given by ( 18.27 ).
Consider next the boundary conditions given in terms of moments and shear The reason is two-fold: it is the weak form on which our FE solution is based; and
forces, which may be denoted as the static boundary conditions. Along the boundary from the natural boundary conditions we will be able to decide on the proper static
and referring to (18.38), we are able to prescribe the bending moment M ••• the boundary conditions. ~
twisting moment Mnm and the shear force V..z, i.e. The equilibrium condition ( 18.27) can be written as
(18.47)
are the static boundary conditions. It is somewhat surprising that, whereas we have
~(aMxx) + ~(aMxy) + ~(aMxy) + ~(aMyy) + q =
fu fu fu ~ ~ fu ~ ~
0 (18.48 )
350 Introduction to the finite element method
FE formulation of plates 351
To obtain the weak form of this differential equation, we multiply by the which can be written as
arbitrary weight function v(x, y) and integrate over the region A, i.e.

v!_ (oM"") dA +I v!_(oMxy) dA +I v!__(oM"Y) dA

I A OX OX A OX Oy , A Oy OX
LG:~ M xx + ::~ Myy + 2 :::Y Mxy) dA
+I v!__(aM yy) dA + I vq dA =0 - r~
J. ( ox
OV (Mxxnx + Mxyny) + ov (Mxynx + Myyny)) d fi'
oy
A oy ay A

To perform an integration by parts, the Green- Gauss theorem given in the

form of ( 5.27) and ( 5.28) is used. This results in
+ f~ vV,. d fi' + L vq dA = 0 ( 18.49)

i.v aM"" nx d!l' - I av aM XX dA + i. v oM xynx d!l'- I av oMxy dA With the definitions ( 18.19) and ( 18.28) we have
J~ OX A OX OX J~ Oy A OX ay
( 18.50)
+i_ voM"Ynyd-2' - I ovoM"YdA +i_ vaM yy n d!l'-f avaM yy dA
J~ ax A Oy OX J~ Oy y A Oy ay

+ L vqdA = 0
Moreover, in ( 18.45) it was shown that

(18.51)
Using (18.6) and (18.7) in the boundary integrals gives

f~
v Vxznx d .zro + f
~
v vyzny d.roz - I -av -
oM-
A OX ax
XX
dA - f -av -
aMxy
A OX Oy
- dA
where the unit vector m is orthogonal to n, i.e. m is tangential to the boundary. With
(18.50) and (18.51), (18.49) reduces to

- -ovaMxyd
-- A - I -
I A Oy ax
ovoMyy
--+ I vqdA =0
A Oy Oy A fA
(Vv)TM dA = i_ (dv M •• + dv M.m) d!l'- i_ vV,. d!l'-
r~ dn dm r~
f A
vq dA

(18.52)
In the boundary integrals, advantage is taken of ( 18.39) and the result becomes
Now, as m is tangential to the boundary, we may take d!l' = dm, i.e.
- f ovoM""dA - f ovoM"Y dA - f ovoM"YdA - f ovoMyydA
A OX ax A ax Oy A Oy OX A Oy Oy
f~ dm
dv
-M.m dfi' = f ~ dm
d
-(vM.m) dfi'- f~
dM-
v- nm d.z
ro

+ f~ v V,. d!l' + L vq dA = 0
dm

f
Repeated use of the Green- Gauss theorem (5.27) and (5.28) yields
=
f ~
d( vM.m)-
f
~
dMnm
v--d!l'
dm
=-
~
vdMnmdro
- - .x
dm

Hence (18.52) takes the form

- i. ~v Mxxnx d!l' + f ~ ~ Mxx dA - i. ~v Mxyny d!l' + f
2
02
v MX)I dA
J~uX AuX J~uX AOYOX

- J. ~v Mxynx d!l'; f 02
v M :x:y dA - J. ov Myyny d!l'
~~ A ~~ ~~ ( 18.53)
02
+f
A oy
~ M yy dA + i_J~ vV,. d.P + f A
vq dA = 0
which is the weak form oftlie equilibrium equation sought. We immediately conclude

I
352 Introduction to the fin ite element method FE formulation of plates 353

that the natural boundary conditions are given by y y

a) b)

M •• ; (18.54) mP

Therefore, it is not the quantities M""' Mnm and V... that determine the static
boundary conditions. Rather, it is M"" and the term V... + dM.m/ dm which matter,
and in this term the value of V,.. or of dMnm / dm is irrelevant; it is only the sum
V... + dM.m/ dm that counts. The proper static boundary conditions ( 18.54) are called
Kirchhoffs boundary conditions and the term V... + dM.m / dm is often called the
effective shear force. As M "'" is the twisting moment per unit length, dM nm/ dm has
the dimension of force per unit length, i.e. [N/ m], just like the shear force V,. •.
Following a proposal by Kelvin and Tait in 1870, it is possible to obtain a physical L--------------------------x
Figure 18.10 Conditions at sharp corners
interpretation of the effective shear force, and this explanation is the one which most
· often appears in textbooks; cf. Timoshenko and Woinowsky-Krieger (1959). unit tangential vector mQ. The tangential vectors are chosen so that the vectors n
and m, after a suitable rigid-body rotation, have the same directions as the x- and
y-axes, respectively. The sharp corner is characterized by the angle a between t.he
18.3 Concentrated shear forces at corners normal vectors nP and nQ. This angle is measured as positive in the counter-clockwtse
direction.
In addition to the vertical shear force per unit length V,.. caused by the vertical shear Considering ( 18.53 ), let us now evaluate the contribution from the shear force
stress cr"., we have seen the existence of the vertical shear force per unit length per unit length dM nml dm, when we pass from point P to point Q. Choosing d.P = dm
dMnm / dm caused by the horizontal shear stress crnm· The shear force distribution we get
dMnml dm, however, possesses some special properties that are of essential importance
when evaluating a plate response. Even experienced engineers may be hesitant about fQ dMnm d.P = fQ dM.m = Ml - M:m (18.57)
these properties and we shall therefore present a detailed discussion of these matters. JP dm P

The first point of interest is that the distribution of the shear force per unit length The exact plate deflection w is a continuous function and so are arbitrary orders of
dMnm / dm along the boundary of the plate comprises a self-equilibrating system of differentiation of the deflection w and due to ( 18.31) we conclude that also the
forces. To see this we integrate dM.m / dm along the entire boundary, and as we may moments vary in a continuous manner. As points P and Q are located infinitely close,
take d.P = dm (cf. Figure 18.8) we obtain we therefore have
p -MQ-M.
M xx - xx -
MPyy -- MQ-M.
yy - YY'
W xy =MQ=xy
M xy ( 18.58 )
J: dMnmd.P=J: dMnmdm=J: dMnm=O (18.55)
XX'

J.s!' dm J.s!' dm J.s!' Using (18.58), (18.39) provides

Vertical equilibrium of the entire plate then requires that Ml = n~m~ Mxx + n~m~Myy + (n~m~ + n~m~)Mxy
f9' V,.. d.P = t q dA (18.56) M:m = n~m~Mxx + n~m~ Myy + (n~m~ + n~m~)Mxy
Inserting (18.59) into (18.57) yields
( 18.59)

i.e. the total vertical load is in equilibrium with the total shear force due to V,.•.
The second point of interest is that the distribution of the shear force per unit
length dM.m/ dm exhibits a discontinuity at sharp corners of the boundary. That is, JPfQ dMnm
dm
d.P = MQ -
nm
MP = V
• nm *
at sharp corners of the boundary we have a concentrated shear force. To illustrate ( 18.60)
this remarkable feature, we consider Figure 18.10(a). At point P on one side of the
corner we have the unit normal vector nP and the unit tangential vector mP. At
point Q on the other side of the corner, we have the unit normal vector nQ and the

l
352 Introduction to the finite element method FE formulation of plates 353

that the natural boundary conditions are given by y y

ol b)

( 18.54)

Therefore, it is not the quantities M""' Mnm and V,. that determine the static
boundary conditions. Rather, it is M"" and the term V,. + dMnm fdm which matter,
and in this term the value of V,. or of dM.m f dm is irrelevant; it is only the sum
V,. + dM.m fdm that counts. The proper static boundary conditions ( 18.54) are called
Kirchhoff's boundary conditions and the term V,. + dM.m fdm is often called the
effective shear force. As Mnm is the twisting moment per unit length, dM.m / dm has
the dimension of force per unit length, i.e. [N/ m], just like the shear force V,•.
Following a proposal by Kelvin and Tait in 1870, it is possible to obtain a physical L--------------------------x L--------------------------x
Figure 18.10 Conditions at sharp corners
interpretation of the effective shear force, and this explanation is the one which most
· often appears in textbooks; cf. Timoshenko and Woinowsky-Krieger (1959). unit tangential vector mQ. The tangential vectors are chosen so that the vectors n
and m, after a suitable rigid-body rotation, have the same directions as the x- and
y-axes, respectively. The sharp corner is characterized by the angle CJ. between t.he
18.3 Concentrated shear forces at corners normal vectors nP and nQ. This angle is measured as positive in the counter-clockwtse
direction.
In addition to the vertical shear force per unit length V,. caused by the vertical shear Considering (18.53), let us now evaluate the contribution from the shear force
stress cr•• , we have seen the existence of the vertical shear force per unit length per unit length dMnm fdm, when we pass from point P to point Q. Choosing d2 = dm
dM.m fdm caused by the horizontal shear stress crnm · The shear force distribution we get
dMnml dm, however, possesses some special properties that are of essential importance
when evaluating a plate response. Even experienced engineers may be hesitant about rQ dMnm d2 = IQ dMnm = M!;2,.- M~m (18.57)
these properties and we shall therefore present a detailed discussion of these matters. JP dm P

The first point of interest is that the distribution of the shear force per unit length The exact plate deflection w is a continuous function and so are arbitrary orders of
dM.m f dm along the boundary of the plate comprises a self-equilibrating system of differentiation of the deflection w and due to (18.31) we conclude that also the
forces. To see this we integrate dM. m/ dm along the entire boundary, and as we may moments vary in a continuous manner. As points P and Q are located infinitely close,
take d2 = dm (cf. Figure 18.8) we obtain we therefore have
(18.58)
!, dMnm d2 =!, dMnm dm =!, dMnm = 0 (18.55)
J.2' dm J.2' dm J
.2' Using ( 18.58), ( 18.39) provides
Vertical equilibrium of the entire plate then requires that M!;2,. = n~m~Mxx+ n~m~Myy + (n~ m~ + n~m~)Mxy
f.2' v,. d2 = Lq dA ( 18.56) M~m = n~m~Mxx + n~m~Myy + (n~m~ + n~m~)Mxy
Inserting ( 18.59) into ( 18.57) yields
( 18.59)

i.e. the total vertical load is in equilibrium with the total shear force due to V, •.
The second point of interest is that the distribution of the shear force per unit
length dM.m/ dm exhibits a discontinuity at sharp corners of the boundary. That is,
at sharp corners of the boundary we have a concentrated shear force. To illustrate
this remarkable feature, we consider Figure 18.10(a). At point P on one side of the
iQ dMnm d..<l'
P dm
= MQnm - MP = V
~"m *
( 18.60)

corner we have the unit normal vector nP and the unit tangential vector mP. At
point Q on the other side of the corner, we have the unit normal vector nQ and the

I
354 Introduction to the finite element method FE formulation of plates 355

A 8

a
Figure 18.1 I Rectangular plate
Figure 18.12 Simply supported rectangular plate exposed to a sinusoidal load q

where the definition V* = M~- M:m has been introduced. From ( 18.60) it follows, see Figure 18.12. We take q0 as a positive quantity; that is, the load q acts in the
as expected, that if the corner is smooth, i.e. nP = nQ and mP = mQ, the distribution negative direction of the z-axis, or downwards.
of the shear force per unit length dM nm/ dm is continuous, i.e. M~ = M:m and V* = 0. The material is assumed to be isotropic and it is straightforward to check that
However, at sharp corners of the boundary we have nP =1= nQ and mP =1= mQ. That is, the solution to (18.33) subjected to the boundary conditions w = 0 and Mnn = 0 is
(18.60) shows that the distribution of the shear force per unit length dMnm fdm exhibits given by
a discontinuity since, in general, we will have M~ =1= M:m, i.e. V* =1= 0. Recalling that 4 3 2
q0 nx ny n Et ( 1 1)
Mnm has the dimension of moment per unit length [N m / m], the difference w=- cx sin-;-sinb where cx= 12 ( 1 - v2 ) a 2 +b 2 ( 18.65)
Me;, - M:m = V* has the dimension of force [N], i.e. the discontinuity of the
distribution of dMnm f dm at sharp corners emerges in the form of a concentrated The reader may also check that trivial but somewhat involved calculations based
vertical force, which is here termed v*. on (18.63) and (18.65) show that
To evaluate this concentrated shear force for the case of most practical interest,
we consider a corner in the form of a right angle, i.e. ex= n / 2; cf. Figure 18.10(b). VA= V 8 = vc = V0 = - 2/3 ( 18.66)
From this figure we conclude that
* * * *
where the positive quantity f3 is given by

( 18.61) qo(l - v)
( 18.67)
f3 = n ab(1 / a 2 + 1fb 2 ) 2
2

Using (18.61) in (18.60) yields Therefore, as the concentrated shear forces given by ( 18.66) are negative, they act in
V* = M~- M:m = 2n~n~(Mxx - Myy) + 2[(n~? - (n~?JMxy ( 18.62) the negative direction of the z-axis, i.e. downwards. Moreover, along the boundaries
AB and DC we get the following shear forces per unit length:
As an illustration, consider the rectangular plate shown in Figure 18.11. At
corner C, we have (nP)T = [1 0], i.e. (18.62) gives V~ = - 2M~y· Proceeding in the V = -f3an- ( -12 + -12 ) sm-·
. nx dMnm n . nx
- - = f3 - sm - (18.68)_
same way for the other corners gives nz 1- v a b a' dm a a
)
and along the boundaries BC and AD we obtain '
v~ = 2M~y ( 18.63)
V = f3bn (_!_+ _!_)sin1tY. dMnm = fJ'!!.sinny (18.69)
nz 1 - v a 2 b 2 a ) dm b b
To evaluate these results further, we consider the particular case of a simply
supported rectangular plate loaded by a distributed load q per unit area, which varies The shear forces per unit length V..z and dMnm f dm given by (18.68) and (18.69) are
sinusoidally over the plate, i.e. positive, i.e. they act in the positive-direction of the z-axis, or upwards. It is recalled
that the forces given by (18.66), (18.68) and (18.69) are forces which act on the plate.
. nx . ny As a specific example, assume that a = 3, b = 2 and v = 0.3. The concentrated
q = -q 0 sm-sm- (18.64)
a b shear forces V* and the shear forces per unit length dMnm f dm are then as shown in

I
356 Introduction to the finite element method FE formulation of p lates 357
z y 0.095q 0 where

I
I
t0.181q 0
( 18.71)
~==========~'------ x
0.181q
0
la.181q 0 :v,

and n is the number of unknowns for the entire plate. At this point we shall leave
Figure 18.13 Self-equilibrating system of forces consisting of V* and dMnm/ dm the meaning of the shape functions N 1 and the nodal values u1 unspecified and just
mention that, whereas N1depends on x andy, i.e. N 1= N 1(x, y), the u1components
are parameters. From (18.70) and (18.28) we obtain

* = Ba
Vw where *
B =VN (18.72)

i.e.
o2N iJ2Nl iJ2N2 iJ2Nn
iJx 2 iJx 2 iJx2 iJx2
iJ2N iJ2 Nl iJ2N2 j)2 Nn
B= ( 18.73 )
0.441q 0 :Vnz i)y2 i)y2 i)y2 iJy2
Figure 18.14 Distribution of V,,
2-
iJ2N /J 2N t o2N
2----2
o2N
2--"
axay axay axay axay
Figure 18.13, whereas the shear forces per unit length V,, are as shown in With the Galerkin method the following expression is used for the arbitrary
Figure 18.14. In accordance with ( 18.55) the system of forces consisting of the weight function v:
concentrated forces V* at the corners and the continuous distribution of dMnm/dm
away from the corners is self-equilibrating, as is also apparent from Figure 18.13. In v= Nc (18.74)
accordance with (18.56) the shear forces per unit length ~.shown in Figure 18.14 As v is arbitrary, the parameters c are also arbitrary. It appears that
balance the external load due to the load q; cf. Figure 18.12. It is recalled that the
* Bc
Vv= (18.75)
forces V*' dMnm / dm and ~. are forces acting on the plate. Therefore, Figure 18.13
shows that if the plate is not anchored at the corners, the corners will tend to rise, and ( 5.8 ) provides
and this tendency is in accordance with observations found in practice.
dv
- = (Vv)Tn = cT(VN)Tn (18.76)
dn
where
18.4 FE formulation
Following this detailed discussion of various aspects of plate theory, it is timely to
establish the FE formulation. As the deflection w is the unknown function in the ( 18.77)
plate theory (cf. ( 18.32) and (18.33)) we already know that at some stage we will
write the approximation for w quite generally as
Moreover, since v = vT (18.74) can be written as
w = Na (18.70)
V = cTNT ( 18.78)

l
358 Introduction to the finite element method
FE formulation of plates 359
frE>E> E-dgE'S
We are now in a position to derive the FE formulation. The weak formulation
of the equilibrium condition is given by (18.53) which holds for arbitrary weight built-in edge
functions v. Inserting (18.75), (18.76) and (18.78) and noting that cT is independent z w=O. dw =0
dn

)-,
of position yields
7"7'"---....J simpl y supportE-d edge

c T[ LBTM dA - t~ (VN)T nMnn d.P + t~ NT( v,. + d::m) d.P

w=O.Mnn =0

L J
X

+ NT qdA =0 Figure 18.15 Examples of boundary conditions

As cT is arbitrary, we conclude that and insertion into ( 18.79) yields

L BTM dA = f.2' (VN)T nM"" d.P - f.2' NT( V,. + d::m) d.P
(L BTfiB dA )a = t~ NT( V,. + d:m) d.P - t~ (VN? nM"" d.P

-L NTqdA
( 18.79) + L NTqdA ( 18.83)

which is the FE formulation sought. To write this expression in a more compact

fashion define the stiffness matrix K, the boundary vector fb and the load vector f1 by
which is an expression for the equilibrium condition for the plate, i.e. (18.79) holds
irrespective of the constitutive assumptions.
It is recalled that the essential, i.e. the kinematic, boundary conditions are given by K= L BTfiBdA

(18.80) fb = t~ NT( v,. + d::m) d.P - t~ (VN)T nMnn d.P ( 18.84)

and the natural, i.e. the static, boundary conditions are r. = L NTqdA

dMnm
vn. +-- ( 18.81) i.e. (18.83) can be written as
dm

Moreover, we emphasize that, similar to the situation for beams, it is not always
Ka = fb + r. I ( 18.85)

possible to divide the boundary into parts where the kinematic boundary conditions The force vector f is as usual defined by
are prescribed and parts where the static boundary conditions are prescribed. Indeed,
it is entirely possible to have kinematic and static boundary conditions prescribed ( 18.86)
along the same part. For instance, along a simply supported boundary we have w = 0
and M "" = 0. This example as well as other typical boundary conditions are illustrated I.e.
in Figure 18.15.
At this point we introduce the constitutive and kinematic assumptions adopted. Ka =f ( 18.87)
From (18.31) and (18.72) we get
The FE formulation for one element follows in the usual manner from the expressions
M = -fiBa (18.82) above and need not be written explicitly.

l
 -,
360 Introduction to the finite element method FE formulation of plates 361

18.4.1 Completeness and compatibility requirements It is emphasized that, whereas the completeness requirement must be fulfilled, it may
be possible to relax the compatibility requirement and to work with non-conforming
When we made the approximation for the deflection w given by (18.70), we did not plate elements.
specify the meaning of the shape functions N; and the nodal values u; . In order to The discussion above has much in common with the similar discussion for beam
be more specific, we shall investigate the convergence criteria that are fulfilled if the elements. However, whereas it is easy to construct conforming beam elements, it turns
completeness and compatibility requirements are met. out to be impossible to establish conforming plate elements when use is made of a
For infinitely small plate elements, the deflection within each element will, in the simple polynomial expression for the approximation of the deflection w ( cf.
limit, be characterized by an arbitrary rigid-body motion superposed by an arbitrary Zienkiewicz, 1977). We conclude the following :
constant curvature. The curvature is determined by the curvature matrix K defined
by (18.17) and we note that a constant curvature is the simplest deflection state that Most plate elements are non-conforming.
create strains. It is concluded that the completeness requirements take the following
form : As mentioned in Chapter 7 (page 93), the convergence criteria for non-conforming
elements are more involved than for conforming elements. We shall not present an
evaluation of these matters in this introductory text, but merely refer to Cook et al.
Completeness (1989), Crisfield (1986), Hughes (1987), Zienkiewicz (1977) and Zienkiewicz and
Taylor ( 1989) for a proper treatment of this topic.
• The approximation for the deflection w must be able to represent an
arbitrary rigid-body motion.
• The approximation for the deflection w must be able to represent an
18.4.2 Two simple non-conforming plate elements
arbitrary constant curvature matrix.
For a presentation of plate elements we shall confine ourselves to two simple plate
elements used in practice. We shall also illustrate that both elements are
An arbitrary rigid-body motion consists of a translation in the z-direction as well as non-conforming.
rotations about the x- andy-axes (cf. Figure 18.1). That is, the approximation for The simplest possible plate element having six degrees of freedom is shown in
the deflection w must at least include the expression a 1 + a2 x + a 3 y. The curvature Figure 18.16. It was suggested by Morley (1971 ), and it consists of a triangle with
matrix rc is given by ( 18.17) and in order to fulfil the completeness criterion, it follows the deflection degrees of freedom u 1 , u 2 , u 3 at the corners and the rotational degrees
that any approximation for the deflection w must take the form of freedom u4 , u 5 , u6 at the midpoint of the sides. The deflection w is approximated by
( 18.89)
w = a 1 + a 2 x + a 3 y + a 4 x 2 + a 5 xy + a6 y 2 (18.88) A comparison with ( 18.88 ) shows that this is the simplest possible approximation
+ possibly other terms that fulfils the completeness requirement. It appears that the second derivatives of w
are constants, i.e. the moments are constant within the element (cf. (18.23)).
To derive the compatibility requirement, advantage is taken of the concept of
To show that the element is non-conforming, consider the arbitrary element
en-continuity discussed in Chapter 9 (page 202). According to (18.84) and (18.72),
the stiffness matrix contains second derivatives of the shape functions. This implies
that C 1 -continuity is required for the shape functions and thus also for the
z
deflection w. We then arrive at the following compatibility requirement:

Compatibility
X
~, ~

• The approximation for the deflection w must be continuous and

possess continuous slopes over the element boundaries.
Figure 18.16 Simplest plate element

I
362 Introduction to the finite element method FE formulation of plates 363
y y
I
A
c----------o 7.8. 9

& - - - - - - - - - - - < 1 4. 5,6

L__ _ _ _ _ __ _ _ _ _ _~----------x
L ___________________ -+-------x
Figure 18.17 Triangular plate element with six degrees of freedom X: 0

Figure 18.19 Rectangular plate element

z
w = a 1 + a2 x + a3 y + a4 x 2 + a5 xy + a6 y 2

X
)-, + a 7 x 3 + a8 x 2 y + a 9 xy 2 + a 10 y 3 + a 11 x 3 y + a 12 xy 3
It appears that this expression is able to model exactly an arbitrary linear variation
of moments within the element.
(18.91)

To show that the element is non-conforming consider, for instance, the element
b oundary AB shown in Figure 18.19. Two nodal points are located on AB, each
having three degrees of freedom. Along AB where x = a, ( 18.91) provides that
Figure 18.18 Rectangular plate element with 12 degrees of freedom
w = Al + Azy + A3yz + A4y3
boundary AB with the nodal points 1, 4 and 2 in Figure 18.17. With w given by ( 18.92)
ow 2
(18.89), we get from (5.8) that - = A 2 + 2A 3y + 3A4 y
oy
dw = (Vw)Tn = (a 2 + 2a4 x + a 5 y)nx + (a 3 + a 5 x + 2a6y)ny where A 1 , A 2 , A 3 and A4 are certain constants. The deflection w and slope owl oy
dn are known at both point A and p oint B. Insertion of these four conditions in ( 18.92)
implies a unique determination of the four parameters A 1 , A 2 , A 3 and A 4 . That is,
Along the boundary AB, we have y = {3 1 x + {3 2 , i.e. (18.89) and the expression above
provide the deflection w varies continuously over common element boundaries.
However, the slope ow I ox does n ot vary continuously over the element boundary
w = A 1 + A 2 x + A3 x 2 AB. To see this, we derive from (18.91) that
( 18.90)
dw
- = B 1 + B2 x
dn -ow = B 1 + Bzy + B3y z + B4Y 3
ox
where A 1 , A 2 , A 3 and B 1 , B 2 are certain constants. Along the element boundary AB, where B 1 , B 2 , B 3 and B 4 are certain parameters. A unique determination of these
three quantities are known: the displacements at nodes 1 and 2 and the slope dwl dn four parameters requires four conditions, but only two conditions a re available: the
at node 4. These three conditions are not sufficient to determine the five parameters value of ow 1ox at point A and its value a t point B. That is, the slope ow I ox in general
A1 , A 2 , A 3 and B 1 , B 2 entering ( 18.90) in a unique manner. It is concluded that both varies discontinuously over the element boundary AB and we conclude that the
the displacement wand the slope dw I dn, in general, vary discontinuously over common rectangular element of Figure 18.18 is non-conforming.
element boundaries, i.e. the triangular element of Figure 18.16 is non-conforming. Research activity on the FE modelling of plates is very intense, not only for
Consider next the rectangular plate element having 12 degrees of freedom as Kirchhoff plate theory, but also for Mindlin- Reissner plate theory that o nly requires
shown in Figure 18.18. It was suggested by Adini and Clough (1961) and Melosh C 0 -continuity and therefore facilitates the establishment of conforming elements. The
(1963 ), and it has four deflection degrees of freedom u 1 , u4 , u 1 , u 10 and eight rotational reader is referred, for instance, to Cook et al. ( 1989 ), Crisfield ( 1986 ), Hughes ( 1987 )
degrees of freedom u 2 , u3 , u 5 , u6 , u8 , u9 , u 11 , u 12 . The deflection is approximated by and Zienkiewicz (1977) for an evaluation of these topics.

I
 Isoparametric hnite elements 365
19 ( -1. 1) ( 1. 1 )

Isoparametric finite elements y

( -1. -1)

Parent domain
( 1. -1)
~' Global doma i n

Figure 19.1 Mapping from parent domain to global domain

It was emphasized in Chapter 7 that the sides of quadrilateral and brick elements
must be parallel to the coordinate axes in order to behave in a compatible manner. of distortion. Indeed, there are limitations, and they arise from the requirement that
This restriction has severe consequences when modelling bodies with arbitrary the mapping should be one-to-one, i.e. unique. Therefore, we require that one point
geometries and we shall now show how this obstable can be avoided. It will turn in the parent domain should correspond to one point in the global domain and vice
out that it is possible to establish even compatible, i.e. conforming, finite elements versa. To achieve this objective, we differentiate (19.1) using the chain rule to obtain
that have curved boundaries. The concept we shall study is isoparametric.finite elements

;~][de]
originally introduced by Taig (1961) and in its general form by Irons (1966a,b).
Consider a mapping (i.e. a transformation) of one region into another region dx] = [;; ( 19.2)
(Figure 19.1). A square region in the e11-coordinate system is bounded by the lines [ dy ay ay
--
d17
e = ± 1 and 17 = ± 1. This region is termed the parent domain. We want to map this ae a,
region into another region defined in the xy-coordinate system. The region in the
xy-plane is called the global domain. It follows that this mapping is described by The matrix J defined by

1 x=x(e.,); y=y(e.,) (19.1) oxae ox]

J=
a, (19.3)
i.e. for every point given by its e. ,-coordinates in the parent domain, there exists a
corresponding point given by its x, y-coordinates in the global domain. [ayae a,ay
It is evident that if the general form of the region in the global domain could be
used as a conforming element, we have achieved the objective mentioned above. It
is called the Jacobian matrix related to the mapping ( 19.1 ). Moreover, the determinant
will turn out that this objective can be met using isoparametric elements. To show
this, we need to consider the mapping technique as well as the manner in which the of J, i.e. det J, is called the Jacobian. For any given values of de and d17, ( 19.2) implies
unknown function is approximated. Moreover, it will turn out that the establishment that dx and dy are uniquely determined. Suppose, however, that dx and dy are given
of the FE equations now requires that integrations are carried out in the parent and that we want to determine the corresponding de- and d17-values. From (19.2)
domain. and (19.3) it follows that

( 19.4)

19.1 Restrictions on the mapping and it is evident that we require det J =f. 0. In the following discussion we shall assume
that
The mapping shown in Figure 19.1 creates a distorted quadrilateral in the global
xy-plane, and it is natural to investigate whether there are limitations on the degree detJ > 0 ( 19.5)

364

I
366 Introduction to the finite element m eth od Isoparametric finite elements 367
When (19.5) is fulfilled, a one-to-one relation exists between the values de, d17 and These element shape functions possess the usual properties, and we have, for instance,
dx, dy, i.e. the mapping is unique, at least locally. It is assumed that when (19.5) is that NH -1, -1) = 1 and NH1, -1) = N1(1, 1) = N1( -1, 1) = 0.
fulfilled, the mapping (19.1) is unique even globally. Expression ( 19.1) indicates the general form of a mapping from the parent domain
It is emphasized that, even when the mapping is unique, this does not necessarily to the global domain. It turns out that a very advantageous mapping is obtained if
imply that we are able to invert (19.1) and obtain explicit solutions in the form the element shape functions are used. Therefore, consider the mapping defined by
e = e(x, y) and '1 = 1'/(X, y). For the mappings related to isoparametric finite elements
such explicit forms are generally not obtainable. X= x(e, 1'/) = NHe. IJ)Xl + N~(e, I'/)X2 + N3(e, I'/)X3 + N~(e, I'/)X4
( 19.7)
Y = y(e, 1'/) = Nl(e, 1'/)Yl + Ni(e, '1)Y2 + N3(e, '7)Y3 + N~(e, '1)Y4
where it is recalled that the x, y-coordinates of the corner points of the quadrilateral
19.2 Four-node isoparametric quadrilateral element are known quantities (cf. Figure 19.2). If we define the following matrices:

In order to introduce the concept of isoparametric finite elements, let us consider the
four-node isoparametric finite element, which takes the form of an arbitrary
quadrilateral. N•(e, 17) = [N1 ( 19.8)
The region in the parent domain is defined by the four corner points shown in
Figure 19.2, and we want to map this region into the arbitrary quadrilateral shown
in the global xy-plane. This quadrilateral is also defined by its four corner points.
the mapping (19.7) may be written in the compact form
These corner points are identified by their global x, y-coordinates, which we assume
are known quantities.
Each of the four corner points in the parent domain may be associated with an ( 19.9)
element shape function, i.e.
Bearing the properties of the element shape functions in mind, it follows from
N1 = Nl(e, 17); N~ = N~(e, 17); N3 = N3(e, 17); N~ = N.(e, 17) (19.7) that
The explicit expressions for these element shape functions are taken as those of the x(-1, - 1)=x 1 ; x(1, -1)=x2; x( 1, 1)=x3; x(-1, l)=x4
four-node rectangular element treated in Chapter 7. Referring to Figure 7.27 and (19.10)
y( - 1, -1) = y 1 ; y(1, - 1) = Y2; y(1, 1) = y3; y( -1, 1) = Y4
(7.110) we then have
i.e. each corner point in the parent domain is mapped into a corner point of the
N1 = He- 1)(11- 1); N~ = - He+ 1)(1'/- 1) quadrilateral. Consider next what happens when e= - 1. With (19.6) and (19.7) we
( 19.6) obtain
N3 = -He+ 1)('7 + 1); N~ = - He - t)(,., + 1)
x( -1, 17) = - H11 - 1)x 1 + !('1 + 1)x 4
(19. 11)
y( - 1, 1'/) = -!<'1 - 1)Y1 + !('7 + 1)y4

(-1 . 1) ( 1. 1 ) i.e. x and y vary linearly with '7 between the corners 1 and 4 of the quadrilateral.
~ 3 Elimination of11 in (19.11) provides a linear relation between x andy. Therefore, the
y straight line e= - 1 between the corner points 1 and 4 in the parent domain is
mapped into a straight line between the corner points 1 and 4 of the quadrilateral.
In general, for e= C, where C is an arbitrary constant, we obtain a linear relation
between x and y. Likewise, for '1 = C, where C is an arbitrary constant, a linear
6------o2
relation between x andy is also obtained. We have therefore proved that the mapping
1-1.-1) ( 1, - 1) defined by (19.9) carries the square defined by its four corner points in the parent
domain into the arbitrary quadrilateral that has straight boundaries and is defined
Pa r ent domai n Gl obal domain
by its four corner points.
Figure 19.2 Mapping into four-node isoparametric quadrilateral element In conclusion, we have the following: corner points in the parent domain are

l
368 Introduction to the finite element method Jsoparametric finite el ements 369
11 in Chapter 15 that for the approximation of the displacement vector u in solid
4 3 mechanics, we adopted the technique of approximating each displacement component
I - ..!_ n-..L
'I - 2
in the same manner as the one applied to the temperature. Here, we can therefore
........ f.............. j... /. f - 11 - 2 confine ourselves to considering the approximation of one unknown function such
I I as the temperature.
""'"i"""" ~ ~
For the four-node isoparametric quadrilateral element shown on the right in
I : :
"""i"'""'f""" 'l"""" 1 2 Figure 19.2, the temperature T within the element is approximated through
I I I '--~=2
2
~=+ T = T( ~, rJ) = N1(~, 17)T1 + N2(~, 11)T2 + N3(~, rJ)T3 + N~(~, 11)T4 (19. 12 )

Parent domain Global doma in where, as usual, T1 . • . T4 are the temperatures at the nodal points ofthe quadrilateral.
Moreover, the element shape functions are again given by (19.6). From ( 19.8), and
Figure 19.3 Mapping of straight lines given by ~ = C or '1 = C in the parent domain into with the standard notation a <T = [T1 T2 T3 T4 ], (19.12) may be written in the
straight lines in the global domain compact form

1] (19.13)

Comparing (19.9) and (19.13), the characteristic feature of an isoparametric

formulation emerges, namely that the same element shape functions are used in the
mapping and in the approximation ('isoparametric' means ' the same parameters').
Moreover, these element shape functions are expressed in the local coordinates~ and '1·
We shall now see the advantage of expressing the element shape functions in
terms of the ~' 11-coordinates. The objective of the isoparametric formulation of the
four-node element is to obtain a conforming element for an arbitrary quadrilateral,
and we shall now demonstrate that this is true. From (19.6) and (19.13 ) it follows
Figure 19.4 Geometrical restriction of the four-node isoparametric quadrilateral element • that we may write the approximation of T as

mapped into corner points in the global domain; the boundaries of the parent domain ( 19.14)
are mapped into the boundaries of the global domain; and any straight line ~ = C
in accordance with (7.107). Along an element boundary of the quadrilateral, we have
or 11 = C is mapped into a straight line in the global domain. These features are
either~ = ± 1 or 11 = ± 1. Assume, for instance, that~= 1, which means that (19.14)
illustrated in Figure 19.3.
As already indicated in the figures, the corner points are the nodal points of the reduces to
element. It is important that, to achieve the properties summarized above, the
numbering of the nodal points should not be arbitrary. In fact, we must require that
nodal numbers at one element boundary in the global domain should also be located This expression, which is a linear relation in 1], is uniquely determined by the two
at one element boundary in the parent domain. This objective is most easily fulfi11ed parameters a 1 + a 2 and a 3 + a 4 . However, along the element boundary the
by labelling the nodal numbering in ascending order in the counter-clockwise temperatures at the two nodal points are known, and these two nodal temperatures
direction. This convention is adopted here, as well as in other textbooks. uniquely determine the two parameters a 1 + a 2 and a 3 + a 4 • Therefore, the
In order to achieve a one-to-one mapping, we previously derived the general temperature variation along an element boundary is uniquely determined by the two
requirement that det J > 0 ( cf. ( 19.5)). In the present case of a four-node quadrilateral, nodal temperatures at this element boundary. This implies that the four-node
it can be shown that this requirement is fulfilled if all internal angles are less than isoparametric element behaves in a conforming manner for an arbitrary quadrilateral.
180°; see Strang and Fix (1973, p. 158). Figure 19.4 illustrates this restriction of the Finally, we shall mention that if two of the nodal points of the quadrilateral are
otherwise arbitrary quadrilateral. chosen to be one and the same point, it is possible to obtain exactly the three-node
Having discussed the mapping technique using the element shape functions;· we triangle treated in Chapter 7. The reader is referred to Hughes ( 1987, pp. 120- 3) for
now turn to the approximation of the unknown function or functions. We have seen details.

I
370 Introduction to the hnite element method Isoparametric hnite el ements 371

19.3 Eight-node isoparametric quadrilateral element are located on the straight line y = ax + fJ, and let us scrutinize the form of the entire
global boundary 1, 8, 4. The coordinates of the three global nodal points are clearly
As the next example, we consider the isoparametric formulation of the eight-node related through ·
serendipity element treated in Chapter 7. The mapping is illustrated in Figure 19.5, Y1 = axl + fJ; Y4 = a:x4 + fJ; Ys = axs + fJ
and as with ( 19.9) this mapping is given by
Inserting these values into ( 19.18) yields
x = x(~, IJ) = N·( ~, rJ)x•; y = y(~, rJ) = N·(~, rJ)y• ( 19.15) y( -1, rJ) = -t(l - rJ)rJaxl + tO+ rJ)rJax4 + (1- ry 2)a:xs + fJ
where x• and y• contain the x- and y-coordinates of the nodes of the element in the and using ( 19.17) results in
xy-plane. The element shape functions are given by (7.120) and (7.121), i.e. y( -1, ti) = etx( -1, ti) + fJ
N~ = - ±(1 - ~)(1 - rJ)(1 + ~ + rJ); N~ = t(l- ~ 2 )(1 - rJ)
That is, if the global nodal points 1, 8, 4 are located on a straight line, the entire
global element boundary 1, 8, 4 is located on that line. In general, however, the global
N].= -t(l + ~)(1 - 1])(1- '+ rJ); N6 =tO+ ~)(1 - rJ 2) element boundary 1, 8, 4 will be curved. As in the derivation of(19.17) and (19.18),
( 19.16) it follows that any straight line in the parent domain given by ~ = C or ti = C, where
N~ = -t(l + ~)(1 + rJ)(1 - ~ -rJ); N; = !(1 - ~ )(1 + rJ)
2

C is an arbitrary constant, niaps into a curved line in the global domain. These
N~ = -i(l - ~)(1 + 11)(1 + ~ - rJ); N"s = t(l- ~)(1 - 11 2 ) features are illustrated in Figure 19.6.
As with the four-node isoparametric element, the numbering of the nodal points
From (19.15) and (19.16), it appears that x( -1, - 1) = x 1 , y( - 1, - 1) = y 1 and in the present case is not arbitrary. In order to achieve the mapping of element
x (O, - 1) = x 5 , y(O, -1) = y 5 , etc., and we therefore obtain the correspondence boundaries in the parent domain into element boundaries in the global domain, we
between nodal points in the parent and global domains indicated in Figure 19.5. must require that nodal points on one element boundary in the global domain are
Let us now investigate how an element boundary in the parent gpmain maps also located on one element boundary in the parent domain. Moreover, in the two
into the global domain. As an example we assume that ~ = -1, and-11 follows from domains, corner nodes must correspond to corner nodes and mid-side nodes must
(19.16) that the only non-zero element shape functions are N~ · N'4 and N8. With correspond to mid-side nodes. These requirements are fulfilled by the numbering
(19.15) we then obtain shown in Figures 19.5 and 19.6.
X( - 1, IJ) = -!(1 -rJ )IJX1 + t(l + IJ)IJX4 + (1 - 1] 2 )x 8 (19.17) In addition to these requirements, we also have the requirement of a one-to-one
mapping, i.e. det J > 0; cf. (19.5). For the eight-node element considered, it can be
y( - 1, 1J) = -t(l - 1J )IJYt + t(l + 1J )IJY4 + (1 - 1J 2 )Ys ( 19.18 ) shown that this requirement is fulfilled if all internal angles are less than 180° and
As both x and y, in general, vary quadratically with IJ, the straight boundary 1, 8, 4 if the mid-side nodes are in the 'middle half' of the distance between adjacent corners,
in the parent domain is mapped into a curve through the global nodal points 1, 8, 4. though the 'middle third' shown in Figure 19.7 is used in practice (see Zienkiewicz
Now let us consider the special situation where the global nodal points l, 8, 4
/
1]
~ 7 3
()---,--{!>--.--()
(-1,1) 7 (1 ,1)
~ o--....:-<Jt>----c 3
2
8
6
8 '-----e-- ~

5
1 0----c>---~ 2
[-1,-1) 5 ( 1,-1)
Figure 19.6 Mapping of straight lines given by ~ = C or 11 = C in the parent domain into
Figure 19.5 Mapping into eight-node isoparametric quadrilateral element curved lines in the global domain

/ I
372 Introduction to the hnite element method Isoparam e tric hnite ele m ents 373
1]

u{ y

L.
y

~. L~ safe zone
for m idpoin t

L/3

Figure 19.8 The mismatch illustrated is not possible

Figure 19.7 Geometrical restrictions on the eight-node isoparametric element
I. 7 3
1] o-----~>------0

and Taylor, 1989, p. 158). After this discussion of various aspects of the mapping, l, r'>---?.:...ffi---<13
6
we now turn to the approximation of the unknown function, for instance the
temperature. As with the four-node element, the temperature T within the
isoparametric element, shown on the right in Figure 19.5, is approximated through 8 2

( 19.19) 0

2
·I
5 X
where the components of Ne are given by ( 19.16) and ae contains the temperatures
at the nodal. points. Again we observe the characteristic feature of the isoparametric I Figure 19.9 Modelling of singularity around crack tip by placing mid-nodes at quarter-points
concept, namely that the same element shape functions are used in the mapping and (
in the approximation (cf. (19.15)). Moreover, these element shape functions are and the geometry are modelled in the same way (cf. (19.15) and (19.19)), it follows
expressed in the local coordinates ~ and '1· that knowledge of the global nodal coordinates determines the global element
The objective of the isoparametric formulation of the eight-node element is to boundaries in a unique manner. This, in turn, implies that two adjacent elements will
obtain a conforming element even for the complicated geometry illustrated in Figure join in a smooth way without any kind of gaps or overlapping. Consequently the
19.5, and we shall now demonstrate that this is true. From (19.16) and (19.19) it mismatch between two adjacent elements illustrated in Figure 19.8 is not possible.
follows that we may write the approximation of T as We have previously mentioned the restriction in selecting the position of the
T = T(~, IJ) = ex 1 + ex 2~ + ex 31J + ex4 ~ + ex 5~IJ + ex6 1} 2 + ex 7 ~ 1} + ex/8~1}
2 2 2
(19.20)
mid-side nodes (cf. Figure 19.7). It is of interest, however, that a violation of this
restriction may result in some appealing new properties of the element. In fact, if two
in accordance with (7.122). Along an element boundary we have either ~ = ± 1 or mid-side nodes are located at the quarter-points, as shown in Figure 19.9, and if
IJ = ± 1. Assume, for instance, that ~ = 1, which means that ( 19.20) reduces to linear elasticity is considered, it can be shown that a singularity arises in the
displacement field and that this singularity is of the same type as that existing for
T = ( ex 1 + ex 2 + ex4 ) + (ex 3 + ex 5 + ex 7 )1J + (ex6 + ex 8 )1} 2 problems in linear elastic fracture mechanics. As established by Henshell and Shaw
This quadratic expression in IJ is uniquely determined once the three parameters (1975) as well as by Barsoum (1976), this interesting property can be used to model
ex1 + ex2 + ex4 , ex 3 + ex 5 + ex 7 and ex 6 + ex8 are identified. Along the element boundary, cracks in an efficient manner. In the case illustrated in Figure 19.9, the crack tip is
the temperatures at three nodal points are known, and these three temperatures located at nodal point 1.
uniquely determine the three parameters ex 1 + cx2 + ex4 , ex 3 + ex 5 + ex 7 and ex6 + ex 8.
Therefore, the temperature variation along an element boundary is uniquely
determined by the three nodal temperatures at this boundary. This implies that the 19.4 ~ree-dimensional isoparametric elements
eight-node isoparametric element behaves in a conforming manner for the arbitrary
geometry illustrated in Figure 19.5. The two-dimensional four- and eight-node isoparametric elements have been discussed
From this observation we may draw a further conclusion. As the temperature above, and by now it should be evident how to formulate other isoparametric elements.

I
374 Introduction to the finite element method lsoparametri c finite el ements 375
follows:

y Isoparametric elements

~)-, The same functions - the element shape functions - are used to describe
the geometry of the element and the approximation of the unknown
function. Moreover, these shape functions are prescribed in terms of the
z e,11-coordinates.
Figure 19.10 Eight-node three-dimensional isoparametric element

This concept may be generalized directly to three dimensions, and with obvious
notation we get

X= x(e, 1'/, 0
= N•(e, 1'f, o x•; y = y(e, 1'f, () = N•(e, 1'/, ()y•;
(19.23)
z = z( e. I'J, 0 = N•( e. I'J, ()z•

and

T = T( e, I'J , () = N•(e, 1'f, ()a• (19.24)

Figure 19.11 Twenty-node three-dimensional isoparametric element I
( For one-dimensional problems we have

Without going into any detail, the three-dimensional eight- and twenty-node X= x(e) = N•(e)x• ( 19.25)
isoparametric elements are illustrated in Figures 19.10 and 19.11, respectively.
and
/
r = r(e) = N•(e)a• (19.26)
19.5 General isoparametric formulation-
convergence requirements For the two-dimensional four- and eight-node isoparametric elements, we proved
above that they fulfil the compatibility requirement and that no mismatch between
For two-dimensional problems, the mapping is obtained through adjacent elements occurs. These conclusions can easily be generalized to all the
elements considered in Chapter 7 and we therefore arrive at the following general
( 19.21) conclusions:

e,
where N•( 11) is the element shape function matrix, whereas x• and y• contain the If an element behaves in a conforming, i.e. compatible, manner in the
x- and y-coordinates of the nodes of the element in the xy-plane. parent domain, its isoparametric version also behaves in a conforming way
The unknown function - the temperature T - is approximated through and no mismatch between adjacent elements exists.

(19.22) As discussed in Chapter 7 (page 93), the convergence requirement is fulfilled if

the compatibili'y and the completeness requirement~ are satisfied. ~twas s~own above
where a• contains the temperatures at the global nodal points. Comparing ( 19.21) that the comp¥J:ibility, or conforming, requirement ts fulfilled, so tt remams to prove
and ( 19.22), the fundamental issue of the isoparametric formulation is evident, as that the completeness requirement is satisfidd. It is recalled that the completeness

I
378 Introduction to the finite element method Jsoparametri c finite elemen ts 379
y
1)
Oxb

dl)L v --:~·-~·C,
y-
0 --- ·- - - /
~-------------------x 1 d~ /

L.___ _ _ _ _ _ _ _ ~
L-------------------x
z Figure 19.13 Transformation of incremental area in parent domain to global domain
Figure 19.12 Cross-product of vectors a and b
b is obtained ford(= 0 and d11 # 0, i.e. (19.36) yields

and this expression for A may be written as

A= ja x bl =Ide{:: ::JI ( 19.35)

Considering two-dimensional elements, we have from ( 19.2) and ( 19.3) that The incremental area dA spanned by the incremental vectors a and b is given by
( 19.35), i.e.
I
[~;] [~~] det[~;oy oy~~] d~
ax ox]
dx] = J[d(J where J = o( ot~ !
[ dy ( 19.36) dA = d( df/ ay x = d( df/ = d11 jdet J j
dt~
[ oy ay -
oy - --
ae at~ ae a11 ae a11
The components of the Jacobian matrix J are easily derived from (19.21), and we obtain According to (19.5), we have that det J > 0, i.e. the following fundamental result is
derived:

OX oN•
-= -x·· -= -x
ox oN• e I dA = d( df/ det J I ( 19.38)
ae ae ' at~ at~
( 19.37) This result is easily generalized to three dimensions ( cf. for instance Zienkiewicz
oy aN· oy aN• and Taylor, 1989), and we obtain
ae = --af y•; at~ = at~ y•
I dV = d( df/ d( de~, I (19.39 )
We shall now determine the relation between incremental areas in the parent domain Here (, 11 and ( are the coordinates in the parent domain and
and the global domain. Consider two straight lines given by ( = C 1 and 11 = C 2 in
the parent domain where C 1 and C 2 are arbitrary constants ,.(see Figure 19.13). ox ox ox
According to the transformation x = x( (, 11) and y = y( (, 11 ), these two straight lines ae d11 d(
map into two lines in the global domain, and these lines will, in general, be curved;
cf. Figure 19.13. At the point defined in the figure by the intersection of lines ( = C 1
"' = oy oy oy ( 19.40)
and 11 = C 2 , we consider in the parent domain the increment d( along 11 = C 2 and ae a11 o(
the increment d11 along ( = C 1 • The mapping (19.36) carries d( into the vector a and oz oz oz
d11 into the vector b. The vector a is obtained for d( # 0 and dt~ = 0, and the vector a~ o11 o(
380 Introduction to the finite element method lsoparametric finite elements 381

In one dimension, we clearly get incremental length d2' in terms of the local variables. From ( 5.17) it follows that

I dx = d~ det J {19.41)
d2 = (dx 2 + dy 2 ) 1 ' 2 ( 19.44)
and it is recalled that d2', by definition, is a positive quantity. Evidently, the boundary
where det J = dxj d~. of the two-dimensional body always coincides with some element boundaries. Along
In the FE method, integrations 'over regions are in practice performed elementwise such an element boundary we have either d~ = 0 or d'7 = 0 ( cf. Figure 19.2). Assuming,
and with the relations (19.38), (19.39) and (19.41), we obtain the followin~ for instance, that d'7 = 0, ( 19.36) yields
transformation of integrals over one-, two- and three-dimensional elements
respectively: ' ax
dx = - d~; dy = -
ay d~ (19.45)
a~ a~

{ f(x)dx= f/(x(~))detJd~ Inserting (19.45) into (19.44) gives

ax)z (ay)z]1'2
t
a

f 1 f/(x(~, '7), y{~,JL2ldet J d~ d'7 d2'= 1 d~l [( -a~ + -a~ ( 19.46)

f(x,y)dA =
a
where oxjo~ and oyjo~ are determined by (19.37). Moreover, as '1 = 1 or '1 = - 1
Laf(x,y, z)dV = f l f t f/(x(~, l'f, (), y(~, '1, (), z{~, l'f, 0) along the element boundary considered, ( 19.46) should be evaluated accordingly.
Therefore, letting f = f(x, y) denote an arbitrary function and letting 5t;, denote the
x det J d~ d'7 dC pertinent boundary of one element, we obtain
(19.42)
In these relations, f denotes an arbitrary function. Moreover, it is emphasized that
La, Aa and V" here refer to the regions occupied by one element in the one-, two- and
three-dimensional case, respectively.
I
( I !E:f(x, y )d2=
fl_/(x(~, l'f),y(~,l'f))
[(ax)z + (oy)z]1'2
a~ 0~ d~ ( 19.47)

where either '1 = 1 or '1 = - 1. An analogous expression is derived if d~ = 0 holds

along the element boundary considered.
For three-dimensional problems, the incremental area dS of the boundary surface
has to be expressed in terms of the local variables. It is emphasized that, in contrast
to dA given by (19.38), dS denotes the incremental area of the boundary surface that
19.6.2 Transformation of boundary integrals is located in the xyz-space. This boundary coincides with some element boundaries.
, Along an element boundary of a three-dimensional element we have d~ = 0 or d'7 = 0
The transformation formulae given in ( 19.42) refer to integrations over the region or dC = 0 (cf. Figure 19.11). Assume, for instance, that dC = 0. As in Figure 19.13,
that i_s occupied. by one element, and as such they are applicable when evaluating we then considertwb straight lines given by ~= cl and '1 = c2in the parent domain
the stiffness matnx K and the load vector f1• However, when determining the boundary where C 1 and C 2 are arbitrary constants. Along '1 = C 2 , we consider the incremental
vector_fb , boundary integrals need to be evaluated (cf. for instance ( 10.15) and ( 10.16)). vector given by d~ =I= 0 and d'7 = 0 (as well as dC = 0). With ( 19.40), this incremental
For one-dimensional problems, the boundary integrals present in the expression vector transforms into the following vector a in the xyz-space:
for the boundary vector fb reduce to boundary terms (cf. (9.13) and (9.14)). With
f(x) being an arbitrary function we clearly have
OX
a~

wher~ (a, b) refer to the region occupied by one element.

(19.43)
·{1\a,d( wher~ at = -
ay
a~
oz
( 19.48)

For two-dimensional problems (cf. for instance ( 10.15)), we need to express the a~

I
382 Introduction to the finite element method lsoparametric fini te elements 383
Likewise, the incremental vector given by d~ = 0 and d17 =f. 0 (as well as d( = 0) is where
considered along~= C 1 . Using (19.40), this incremental vector transforms into the
following vector bin the xyz-space:
ox
01]
8y
~= -I.!l'ncc
N •Tht d..<t' - I
~2
N•Tq.t d..<t' ( 19.53)

bl =
01]
8z
(19.49)
{f = -IA.
NeTQt d A

01]
and
We emphasize that the element boundary was assumed to be given by d( = 0. More
specifically, we have either ( = - 1 or ( = 1 (cf. Figure 19.11) and the vectors a 1 and aN·]
b 1 , given by (19.48) and (19.49) should be evaluated accordingly. Be = OX (19.54 )
We are now in a position to derive the result sought. From (19.34), (19.48) and [ aN·
(19.49) it follows that oy
(19.50)
Tlte, element boundary vector ~ and the element load vector :£f present no
problems. The element shape function matrix N• is given in terms of ~ and 1], and
With f = f(x, y, z) being an arbitrary function, we can then conclude that
I the remaining quantities entering the integrands of~ and :£f can also be expressed
( in the local variables ~ and 17 via the relations x = x( ~. '1) and y = y( ~. 17 ). When the

L. f(x, y, z) dS = f f /(x(~.
1 1], (), y(~. 1], 0. z(~. 1], m
entire integrands of~ and :£f have been expressed in terms of ~ and 1], the integratibn
limits are changed in accordance with (19.47) and (19.42).
However, the evaluation of the element stiffness matrix K• presents a new problem.
x I a 1 x b 1 1 d~ d17 ( 19.51)' The point is that the B•-matrix is obtained by differentiating N• with respect to x
andy (cf. (19.54)), but for an isoparametric formulation, the N •-matrix is given in
where either ( = 1 or ( = - 1 and S refers to the pertinent boundary of one element. terms of the ~. 17-coordinates and not the x, ..,v-coordinates.
Analogous expressions are derived if d~ = 0 or d17 = 0 holds along the element In order to determine the derivatives ofN• as given by ( 19.54 ), we first differentia te
boundary considered. N• with respect to ~ and 11· It follows that

aN•]
a~ = [oN•
ax ox
a~
+ oN•
oy 8y] [ox
a~ = a~ a~
8y][8N •]
ax
19.7 Evaluation of FE equations using isoparametric aN·
[-
aN· ax aN· ay
--+ - -
ax
-- -
oy oN•
elements 811 ax a17 ay a17 a17 a17 ay
. /-
The use of isoparametric elements implies that in order to evaluate the F E equations, Use of ( 19.3 ) g1ves
some slight reformulations are needed. T o see this, consider as an example the problem

[ a~·] = JT[a~·] (
of two-dimensional heat flow. From (10.39)-( 10.42) we have
19.55 )
K•a• = ~ + :£f ( 19.52) aN· aN·
. 01]
- -8y '

I
384 Introduction to the finite element method lsoparametric finite elements 385
I.e. Use of these values for oN'[ I ax and oN'[ I oy in ( 19.58) means that all the components
of the B•-matrix can be determined. However, contrary to the heat flow problem ( cf.

Be = aN·]
ax = ( JT) - 1
[aN·]
a~
(19.56)), it is not possible to derive a simple expression for B• and, consequently, a
simple expression for the element stiffness matrix K• - similar to (19.57)- cannot be
[ay
-
aN· aN·
-
ay
(19.56)
established.

Using this expression in (19.53) and changing the integration limits in accordance
with (19.42) yield 19.7.1 Example

In order to illustrate the mapping technique for isoparametric elements, we shall

consider a special mapping by which the Jacobian matrix J becomes a constant matrix.
(19.57) For this wrpose we evaluate the four-node isoparametric element when it takes
the form of an"arbitrarily located parallelogram; see F igure 19.14. In this case we
have the conditions

The constitutive matrix D and the thickness t are in general functions of x and y.
Using x = x(~, 17) andy = y(~, 17), both D and t can be expressed directly in terms
of~ and '7 and the integration of (19.57) can then be carried out.
Therefore, since x 4 = x 1 - x2 + x 3 and y4 = y1 - y2 + y 3 , it follows from (19.6) and
( 19.9) that
It is evident that if the Jacobian matrix J is not a constant matrix, the inverse
matrix J- depends in a highly complicated manner on~ and '1· Consequently, an
1
x = -!(~- l)x 1 + !(~- 17)x2 + t('7 + 1)x3
exact analytical integration of ( 19.57) is in general not possible and instead one has I
( Y = -!(~- l)y1 + t(~- '1)Yz + !('7 + l)y3
to resort to approximate integration techniques, the so-called numerical integrations.
Such techniques are discussed in detail in the next chapter. The Jacobian matrix J is given by (19.3), and we obtain
For elasticity problems, the expression for the B•-matrix is considerably more
complicated than ( 19.56) and this implies also that the expression for the element J = [!(xz - xl) !(x3 - Xz)J (19.60)
stiffness matrix K• becomes complex. To see this, consider for instance two- !(Yz - Yt) !(Y3 - Yz)
dimensional elasticity for which we have It appears that J is a constant matrix, and it is easily shown that
oN! 0
oNi 0
aN~ A
0 det J =-
ax ax ax 4 I
B•=
oN! 0 aNi
0 aN~
~
where A is the area of the parallelogram ( cf. Figure 19.14). This result is in accordance I
0 ( 19.58)
ay ay ay
I

oN! oN1 oNi oNi

-- -- - - --
ay ax ay ox
aN•ne aN·n,
ay ax l]
!
3
t. 3
where n. is the number of nodal points for the element ( cf. ( 15.11 )). By analogy with
(19.55), we find that

- -
aNi]
ax _ JT _ 1 [oN'[]
a~

[a~; - l ) a~; (19.59)

2'
Figure 19.14 Four-node isoparametric element in the form of an arbitrary parallelogram

J
386 Introduction to the fini te element method

with ( 19.38), which, when det J is a constant, yields 20

f J~
A = det J
1 1
de d11 = 4 det J
However, the important point is that when J is a constant matrix, the integrations
necessary to obtain the FE equations for the four-node isoparametric element are
precisely those integrations necessary to obtain the FE equations for the four-node
[Numerical integration
I
element treated in Chapter 7 (cf. Figure 7.27). To show this, consider the element
stiffness matrix K• for two-dimensional heat flow. If the element formulation of Figure
7.27 is adopted, we obtain from (19.53)
\

Ke -_ JY•fx2[oN•T oN•TJ D [o~•J tdxdy

- - --
y, x, ax oy oN• (19.61) We have seen that the use of isoparametric elements forces us to perform the
oy integrations required to obtain the FE equations in an approximate manner, i.e. by
numerical integration. However, the existence of isoparametric elements is not the
where the expression for n• given by (19.54) has been used. In the present case where only motivation for the use of numerical integration techniques. Consider, as an
J is a constant matrix, a comparison of( 19.61) with ( 19.57) reveals that the integrations example, an inhomogeneous material for which the constitutive matrix D depends
?f the ~wo expressions are completely similar. This means that exact analytical on the coordinates, i.e. D = D(x, y, z). If this expression is complex, then even when
mtegratwns can be carried out also for the isoparametric formulation. However, the elements of Chapter 7 are used, it may not be possible to perform the necessary
w~en the four-node isoparametric element differs from the parallelogram shown in integrations in an exact analytical manner. In addition, even though an exact analytical
Figure 19.14, the Jacobian matrix J is not a constant matrix and the inverse matrix I integration may be possible, it may be so complicated that it hampers the establishment
J - will therefor~ take a hi~hly c~mplex form. This implies that expression ( 19.57)
1
( of an efficient FE program. Surprisingly, we shall also see that, while an approximation
for the.element stiffness matnx K• IS not amenable to exact analytical integration. is certainly related to numerical integration, this approximation may, in fact, improve
It Is observed that the arbitrarily located parallelogram treated above may take the FE results. /
the simplified form of a rectangle. In the following, we will first present some basic facts on numerical integration,
and for a more comprehensive treatment, the reader is referred to Dahlquist and
Bjorck ( 1974 ), Froberg ( 1965) and Press et al. ( 1986). Then we will discuss various
aspects of numerical integration in relation to isoparametric elements.
19.8 Need for numerical integrations In order to derive suitable numerical integration techniques, it suffices to consider
the problem
The exa~ple just investigated provided the information that, apart from some simple
geometncal configurations ofisoparametric elements, the Jacobian matrix J and thus
1 (20.1)
J - ar.e functions of the .local variables. This means that the integrations required
to achieve the ~E equatiOns become so complex that exact analytical integration
cannot be obtamed. Therefore, it is necessary to perform this integration in an where /(e) is an arbitrary function and I is the quantity to be determined. The
approximate manner, i.e. numerical integration techniques are called for. This is the problem of solving (20.1) can be viewed as determining the area below the /(~)­
subject of the next chapter. · function, and methods of solving (20.1) are therefore also termed quadrature formulae
F~nall~, w_e may refer the reader to the textbooks by Bathe (1982), Hughes (1987) ('quadrature' means 'area'); see Figure 20.1(a).
and Zienktewicz and Taylor ( 1989) for further information on isoparametric elements.
In these books, the isoparametric formulation of triangular and tetrahedral elements
e
Quite generally, we may select some points ~i in the interval -1 ~ ~ 1, and
these points are termed integration points. We then have
is also treated.
n
I= L f( ~;)H; + R (20.2)
i= 1

387

i
386 Introduction to the finite element method

with ( 19.38), which, when det J is a constant, yields 20 I

1

f
A = det J
1
f~ 1
d¢ d11 = 4 det J

However, the important point is that when J is a constant matrix, the integrations
necessary to obtain the FE equations for the four-node isoparametric element are
precisely those integrations necessary to obtain the FE equations for the four-node
[ Numerical integration
I
element treated in Chapter 7 ( cf. Figure 7.27). To show this, consider the element
stiffness matrix K• for two-dimensional heat flow.lf the element formulation of Figure
7.27 is adopted, we obtain from (19.53)

e _
K -
fY•fx [oN•T
2

---- D
oN•] tdxdy
oN•T] a; (19.61) We have seen that the use of isoparametric elements forces us to perform the
y, x, ox oy [oN·
oy
integrations required to obtain t~FE equations in an approximate manner, i.e. by
numerical integration. However, the existence of isoparametric elements is not the
where the expression for B• given by (19.54) has been used. In the present case where only motivation for the use of numerical integration techniques. Consider, as an
J is a constant matrix, a comparison of( 19.61) with ( 19.57) reveals that the integrations example, an inhomogeneous material for which the constitutive matrix D depends
of the two expressions are completely similar. This means that exact analytical on the coordinates, i.e. D = D(x, y, z). If this expression is complex, then even when
integrations can be carried out also for the isoparametric formulation. However, the elements of Chapter 7 are used, it may not be possible to perform the necessary
when the four-node isoparametric element differs from the parallelogram shown in integrations in an exact analytical manner. In addition, even though an exact analytical
I integration may be possible, it may be so complicated that it hampers the establishment
Figure 19.14, the Jacobian matrix J is not a constant matrix and the inverse matrix
J- 1 will therefore take a highly complex form. This implies that expression ( 19.57) ( of an efficient FE program. Surprisingly, we shall also see that, while an approximation
for the element stiffness matrix K• is not amenable to exact analytical integration. is certainly related to numerical integration, this approximation may, in fact, improve
It is observed that the arbitrarily located parallelogram treated above may take the FE results.
the simplified form of a rectangle. In the following, we will first present some basic facts on numerical integration,
and for a more comprehensive treatment, the reader is referred to Dahlquist and
Bjorck ( 1974 ), Froberg ( 1965) and Press et al. ( 1986). Then we will discuss various
aspects of numerical integration in relation to isoparametric elements.
19.8 Need for numerical integrations In order to derive suitable numerical integration techniques, it suffices to consider
the problem
The example just investigated provided the information that, apart from some simple
geometrical configurations of isoparametric elements, the Jacobian matrix J and thus (20.1)
J - 1 are functions of the local variables. This means that the integrations required
to achieve the FE equations become so complex that exact analytical integration
cannot be obtained. Therefore, it is necessary to perform this integration in an where f( ¢) is an arbitrary function and I is the quantity to be determined. The
approximate manner, i.e. numerical integration techniques are called for. This is the problem of solving (20.1) can be viewed as determining the area below the f(¢)-
subject of the next chapter. · function, and methods of solving ( 20.1) are therefore also termed quadrature formulae
Finally, we may refer the reader to the textbooks by Bathe ( 1982), Hughes ( 1987) ('quadrature' means 'area'); see Figure 20.1(a).
and Zienkiewicz and Taylor ( 1989) for further information on isoparametric elements. Quite generally, we may select some points ¢i in the interval -1 ::; ¢ ::; 1, and
In these books, the isoparametric formulation of triangular and tetrahedral elements these points are termed integration points. We then have
is also treated. n
I= l:J(OHi + R (20.2)
i =1 '

387

I
388 Introduction to the finite element method Numerical integration 389
ol f (~) b) f ( ~}
where m is the number of integration points in the 11-direction. It is emphasized that
the number of integration points in the ~- and tT-directions may differ, i.e. we have
in general that n =I m. Combining (20.6) and (20.7) yields
m n

I~ L L f( ~i• tT)H;Hj (20.8)

-~~>...:>IL<~.LL£.-- ~
j=1 i=1
-1
~I ~2 This expression clearly demonstrates that we can focus our interest on the
HI I' H2
one-dimensional problem given by (20.3).

Figure 20.1 (a) Exact integration; (b) numerical integration

./
where n is the number of integration points, Hi denotes some parameters - or weights
- related to each integration point and R is the so-called remainder. The objective is 20.1 Newton-Cotes quadrature
to make the remainder R as small as possible so that we may write, with close accuracy,
The most obvious numerical integration is obtained by a priori selecting the position
n
of the integration points. Moreover, in order to carry out the integration it is natural
I~ Lf(eJHi (20.3) to approximate the function f( 0 by an easily integrated function that in practice
i= 1
takes the form of a polynomial.
as illustrated in Figure 20.1 (b). As the function f(~) is evaluated at the integration points~;, the approximating
The extension to multiple integrals is straightforward, and to illustrate this we I polynomial is conveniently constructed using the Lagrange interpolation function
consider ( 1? - 1 (e); cf. (7.78) and Figure 7.18. With n denoting the number of integration points,
the Lagrange polynomial 1?- 1 ( ~) is given by

(20.4) ~~ - 1(~) = (~- ~1}(~- ~2) ··· (~- ~i - 1}(~- ~i+tl··· (~ - ~n)
(20.9)
(~; - ~tl(~; - ~2) · ··(~i- ~; - tl(~i- ~i + l) .. ·(~i - ~n)
1

It follows that and it appears that 1? - 1 ( ~) is a polynomial of the order n - 1. In accordance with
Figure 7.18, it is easily checked that 1? - 1 ( ~i) = 1 and l'f- 1 ( ~) = 0 for j =1 i. Therefore,
I~ f (J1 f(~i•'1)H;)d'1
1
(20.5) we approximate f( ~ ) by
n

I<~>~ 2:
i= l? -
1
where n is the number of integration points in the ~-direction. Defining the function <~>f<~i> (20.10)
1
g(tT) by
n
which implies that the approximating polynomial takes the value f( ~i) for~= ~i·
g(tT) = L f(~;, 11)H; (20.6) It follows that
i=1
(20.5) may be written as I = J~1 f(~) d~ ~ f1 Ct1 17 - 1 (~)!(~;)) d~ = it1 (f(~;) L1 l?- 1 (~) d~)
I~ f1 g(tT) dtT
(20.11)
A comparison with (20.3) shows that the weights Hi are given by
i.e.
m

I~ L g(tT)Hj (20.12)
(20.7)
j= 1

I
390 Introduction to the finite element method Numerical integration 391

We have now arrived at the Newton- Cotes integrationformula. It is obvious that 20.1.2 Example 2

For n integration points, Newton- Cotes integration provides an exact Assume that three integration points ~ 1, ~2 and ~ 3 are chosen. We therefore have
integration of a polynomial of the order n - 1. n = 3, and (20.9) provides

The weights determined by (20.12) are given in many textbooks; see for instance q- 1(~ ) = li(~) = (~ ; ~2 )( ~ - ~3)
Froberg (1965). Let us consider two simple examples in order to illustrate the (~1 - ~2)(~1- ~3)
procedure.
~~-1(0 =I~(~)= (~- ~1)(~ - ~3)
( ~2- ~d(~2 - ~3)
.20.1.1 Example 1
13-1(~) =I~(~)= ( ~- ~d (~ - ~2)
Assume that two integration points ~ 1 and ~ 2 are chosen. We therefore haven= 2, (~3- ~1 )(~3- ~2)
and (20.9) provides Choosing the position of the integration points as

z1- 1 (~) = n(~) = ~~-=._ ~; ~~- 1 (~) = zHo = ~~- ~1 ~~=-1; ~2=0; ~3=1
1 2 2- 1
we obtain from (20.12) that
Moreover, we choose the position of the integration points as
~~=-1 ; ~2=1
This implies that
From (20.12) we then obtain
H 1 =H 2 =1
I f~l f(~)d~ ~ f( -1) t + /(0) 1 + /(1) t
X X X

i.e.
I which is the well-known Simpson's formula; cf. Figure 20.3.
flf(Od~ ~ f(~dH1 + f(~2)H2 = f( -1) X 1 + /(1) X 1

This result is illustrated in Figure 20.2( a). Moreover, the above result may be written as
20.2 Gauss integration

f1 /(~) d~ ~ 2t[f( -1) + /(1)]

In Newton- Cotes integration, the position of the integration points was chosen
a priori, and the corresponding weights were then determined so that a polynomial
of the order n - 1 could be integrated exactly by using n integration points.
which shows that we have arrived at the well-known trapezoidal rule; cf. Figure 20.2(b ). Another appealing strategy arises if not only the weights, but also the positions
t

'
f ( ~) f (f l
ol b)
• f (1l

f ( -1 }
f ( -1 )

-1 -1 - 1
I· , ·I I~ .I
Figure 20.2 (a) Newton- .Cotes integration with two integration points; (b) equivalence with
trapezoidal rule Figure 20.3 Newton- Cotes rintegration with three integration points; Simpson's formula

I
392 Introduction to the finite element method Numeri cal in tegration 393

of the integration points are determined so that a given polynomial is integrated Table 20.1 Positions of Gauss points ~'and corresponding weights H ;
exactly. This approach leads to the Gauss integration scheme (occasionally termed J~J(~)d~ = !:7= J(~,) H;
Gauss- Legendre integration), and the positions of the integration points derived are
termed Gauss points.
For th!s purpose consider the following polynomial of order 2n- 1: n= 1
~~~~~ ~~~~~
g(~) = OC1 + OC2~ + OC3~ + ... + or:2n - le• - 2 + or:2n~ 2 n - l
2
(20.13)
n=2
We find that this polynomial contains 2n terms. Integration gives ± 0.577 350 269 189 626 1.~ 000 000 000 000
n=3
I= Jl
- 1
g(~) d~ = 2or:l + ~OC3
3
+
2
... + - -or:2n-1
2n- 1
(20.14) 0.000 ~ 000 000 000
± 0.774 596 669 241483
Q888888888888889
0.555 555 555 555 556
n=4
Assume that n integration points are adopted. The general integration scheme (20.3) ± 0.339 981 043 584 856 0.652145154 862 546
then provides ± 0.861136 311594053 0.347 854 845 137 454

I= fl g(~)d~ Jl g(~i)Hi = (20.15) 0.000 000 000 000 000

± 0.538 469 310 105 683
n=5
0.568 888 888 888 889
0.478 628 670499 366
± 0.906 179 845 938 664 0.236 926 885 056189
With g(~) given by (20.13), we obtain
n=6
n n n
. ±0.238 619 186 083197 0.467 913 934 572 691
I =or: I I Hi+ a2 I ~iHi + or:3 2: ~r H, + ... ±0.661209 386466 265 0.360 761 573 048139
i= I i= 1 i= 1
n n ±0.932469 514 203152 0.171324492379170
+ or:2n - l L
i= 1
a· - l H; + or:2n L
i= I
a·-!Hi I
(
A comparison with (20.14) implies the following conditions:
The positions of the Gauss points and the corresponding weights are tabulated
n n n 2 in many textbooks on numerical analysis and FE techniques. Some useful results
L Hi = 2; L ~iHi = 0; L (!·H.=-·
i= 1 i= 1 i= 1 ' ' 3' taken from Zienkiewicz and Taylor (1989) are shown in Table 20.1.
(20.16) The following two examples illustrate how the weights and the positions of the
n 2 n
L ~~n - 2H·=--· L a•-IH;=O Gauss points can be determined.
i= 1 ' ' 2n- 1' i= 1

As (20.13) contains 2n terms, (20.16) provides 2n conditions. Moreover, in (20.15)

we have 2n unknowns, namely n positions ~i and n weights Hi. That is, the conditions 20.2.1 Example 3
given by (20.16) provide sufficient information for the identification of the Gauss
points ~i and the weights Hi. Assume that two Gauss points are adopted, i.e. n = 2. The conditions in (20.16) then
We therefore conclude that reduce to
I
For n integration points, Gauss integration provides an exact HI + H2 = 2; ~ 1Hl + ~2H2 = 0 ; ~i Hl + ~~ H2 = t; ~i Hl + ~~ H2 = 0 I

integration of a polynomial of the order 2n - 1. We could solve these equations directly, but there is no loss of generality in observing
that symmetry reasons imply that H 1 = H 2 and~~= -~ 2 . With this observation
It follows that for a given number of integration points, Gauss integration provides the conditions above can easily be solved to obtain
the exact integration of a polynomial of a higher order than that obtained by
Newton- Cotes integration. In practice, therefore, Gauss integration is used almost H1 = H2 = 1; ~1 = - 1/fi; ~2 = 1/ fi
exclusively within isoparametric FE formulations. in accordance with Table 20.1.

l
 394 Introduction to the finite element method N umeri cal integration 395

20.2.2 Example 4

Consider next the integration scheme when three Gauss points are used, i.e. n = 3.
From (20.16) it follows that
H1.+ H2 + H3 = 2; e1H1 + e2H2 + ~3H3 = 0;
~iH1 + ~~H2 + ~·~H3 = ~
Figure 20.4 Locations of Gauss points for 1 x l , 2 x 2 and 3 x 3 point integration in parent
~I H1 + e~H2 + e~H3 = 0; e1Hl + ~~H2 + ~jH3 = i; domain

eiHl + e~H2 + e~H3 = o

Based on the results of Table 20.1 and with crosses indicating the positions of
Forreasonsofsymmetry,wehave~ 1 = -e 3 ,~ 2 = OandH 1 = H 3 .1tthenfollowsthat
the Gauss points, the 1 x 1, 2 x 2 and 3 x 3 point integration~ are illustra~ed in the
e~=-Jt; e 2 =0; e3=.Jt and H 1 = ~; H 2 =~; H 3 =~ parent domain in Figure 20.4. These locations apply whether the 1so~aramet~1c el~ment
is distorted or not. However, it should be realized that the Jacobmn matnx J IS not
in accordance with Table 20.1. constant when the isoparametric elements are distorted. In this situation, the
establishment of the K•-matrix implies integrations of complex functions of~ and 11
that are not integrated exactly when a 2 x 2 and a 3 x 3 point integration are adopted
for the four-node and eight-node isoparametric elements, respectively. This suggests
20.3 Introductory remarks on the order of Gauss that when we use isoparametric elements, it is advantageous to adopt an FE mesh
integration of isoparametric elements that distorts the elements as little as possible.
In general, it appears that numerical integration introduces a~ additional
With the choice of the Gauss method for integrating isoparametric finite elements, I approximation into the FE method, and therefore we may expect a h1~h order of
one question remains: namely, the number of Gauss points, i.e. the order of Gauss r integration to be preferable. In practice, this is not the case, and the reason IS tw?-fold:
integration. The number of Gauss points is usually the same in the directions of the (a) numerical integration requires a substantial computational effort that evide~tly
local, e-, 11- and C-axes, but apart from that no precise answer can be given. increases with the order of integration; and (b) the approximation related to numencal
However, in order to provide some guidance on this delicate problem, we may integration may, in fact, improve the FE results, which suggests that a relatively low
ask whether situations exist where the numerical integration results in an exact order of integration should be adopted.
integration. For this purpose consider Figure 19.14. In this figure, the four-node This latter point is somewhat surprising, but it hinges on the fact that exact
isoparametric element takes the form of an arbitrarily located parallelogram. As a integration implies that the stiffness of the body is overestimated when compa:ed
consequence, it was shown that the Jacobian matrix J becomes a constant matrix with the true stiffness of the body, as explained below. Moreover, when the numencal
(cf. (19.60)). According to (19.56) and (19.6) the components of the B"-matrix are integration does not provide the exact result, the corresponding. approximation te?ds
e
then linear functions of and 11· Moreover, if the constitutive matrix D and the to compensate for the otherwise too large stiffness. Hence, maccurate numencal
thickness t are assumed to be constants, it follows from (19.57) that the element integration may provide a more accurate FE solution. . . .
It is emphasized that such inaccurate integration may also mvolve cer~am ~1tfalls
f stiffness matrix K" is obtained as integrations of polynomials of the order of two.
Consequently, when the four-node isoparametric element takes the form of an in terms of so-called spurious zero-energy modes, which we shall consider m the
arbitrarily located parallelogram, a so-called 2 x 2 point integration (i.e. two Gauss following section. It turns out that the only sure way to avoid such pitfalls is to adopt
points in the ~-direction and two Gauss points in the 11-direction) implies that the an integration of a sufficiently high order. . .
numerical integration provides the exact result. We shall now give a heuristic proof that the stiffness of the body IS overestimated
Considering the eight-node isoparametric element, it can easily be shown that by the FE method when exact integration is used. A concise proof may be found,
when it takes the form of a rectangle, the Jacobian matrix J becomes constant. It for instance, in Hughes (1987, p. 187). We consider an elastic body, and we note that
follows that the B•-matrix contains quadratic polynomials in eand 11, and in order the FE solution permits only those displacements that can be represented by the
to obtain the element stiffness matrix K", integrations have to be performed of adopted shape functions. In general, the exact displacement field .will differ from .the
polynomials of the fourth order in e and 11· Therefore, a 3 X 3 point integration assumed displacements. Effectively, the assumed displacement field Imposes constramts
suffices to provide the exact integration result. that prevent the structure from deforming the way it wants to. That is, the FE method

'
396 Introduction to the finite element method N umerical in tegration 397
provides constraints on the displacements and, in effect, the FE approach creates a b)
substitute structure that is stiffer than the real one. Similar arguments hold for heat -,
flow in a body. We conclude that

Exact integrations provide a structure that is too stiff.

[[] J
I
I

Figure 20.5 Rigid-body motions: (a) translation in the direction of the y-axis; (b) translation
in the direction of the x-axis; (c) rotation

20.4 Reduced integration and spurious zero-energy a) 4 b)

__,
modes y
l..-----·1
~+· -
When the Jacobian matrix J is constant and when the order of integration is such
that it yields the exact result, we speak of full integration. Examples are 2 x 2 and
3 x 3 integration for the undistorted four-node and eight-node isoparametric elements,
L, I
L
2
\
~
r--
---- i .........
......... -...j

respectively. Reduced integration means that the order of integration is lower than Figure 20.6 Spurious zero-energy modes for four-node isoparametric element with 1 x 1 point
that of full integration. integration
Before investigating the implications of reduced integration, we note from (13.10)
that the strain energy U of a three-dimensional elastic body is given by
rigid-body motions may create zero strain energy. Such displacement modes are
termed spurious zero-energy modes.
(20.17) / In order to illustrate such spurious zero-energy modes, we next consider the .
! four-node isoparametric element used in two-dimensional solid mechanics. A 1 x 1
point integration, i.e. a reduced integration, is adopted; cf. Figure 20.4. In this case,
where W is the strain energy per unit volume. With

8 = Ba and K= t BTDBT d V
it follows that the element stiffness matrix K• takes the form

where the coefficient ex depends on the weight and the determinant of the Jacobian
(20.20)

it follows that the strain energy of the body can be written as matrix. Moreover, the position of the Gauss point is given by~= rr = 0; cf. Table 20.1
U = taTKa and Figure 20.4. As usual, we have 8 = B•a•, i.e.
(20.18)
The same expression can easily be shown to apply also to one- and two-dimensional (20.21)
bodies.
Consider the four-node isoparametric element when it takes the form of a
As the stiffness matrix K is positive semi-definite (cf. (16.30)), it follows that rectangle; see Figure 20.6. Suppose that the nodal points of this element are displaced
u~o I (20.19)
so that
a•T = [ -C 0 C 0 -C 0 C OJ (20.22)
where the equality sign holds only for rigid-body motions; cf. the discussion of ( 16.30). where Cis an arbitrary constant. The displacements given by (20.22) are illustrated
The rigid-body motions of a rectangular element are illustrated in Figure 20.5, and in Figure 20.6( a). From ( 15.5) it follows that
the corresponding displacement modes are termed zero-energy modes, since no elastic
energy is created by these modes. It is emphasized that the above observation - that Ux = ( - NJ. + N~- Nj + N~)C; Uy =0
U = 0 holds if, and only if, rigid-body motions are considered - is only true when
which, with ( 19.6), reduce to
the stiffness matrix is derived by exact integrations. When reduced integration is
adopted, it will turn out that other displacement modes than those corresponding to ux= - C~rr; uy=O (20.23)
398 Introduction to the finite element method
Numerical in tegration 399
In order to derive the strains, the Jacobian matrix J has to be derived. Referring
to Figure 19.14 and (19.60), we have stiffness matrix, a model that is too stiff is created, and in this respect reduced
integration may be beneficial to the accuracy of the FE solution. In light of these
J = [~(x 2 - xd !(x3 - x2)] facts, we may draw the following conclusions :
. 2(Y2- Yd I(Y3- Y2)
Reduced integration may result in spurious zero-energy modes that
In the present case, the element boundaries are parallel to the coordinate axis, i.e. destroy the FE solution. If spurious zero-energy modes are not created,
x 3 = x 2 and Y2 = y 1 , which leads to reduced integration may increase the accuracy of the FE solution, since
it tends to soften the stiffness of the model.
J=[t(x 2 -x 1 ) 1. 0 ];
0 2(YJ- Y1)
Having considered reduced integration of one four-node isoparametric element,
i.e. (Jr) - 1 = 2
(x2- x1)(y3 - Yd
[Y3 -0 Y1 0J
x 2 - x1
(20.24)
we shall now turn to the identification of spurious zero-energy modes for an FE
mesh of arbitrary elements and with an arbitrary number of Gauss points. It turns
out that it is relatively easy to establish an expression for the number of spurious
As in ( 19.59) we have zero-energy modes that at least must be present, but it is more difficult to identify
the exact number of spurious zero-energy modes. We shall therefore concentrate on
oux] [oux]
OX =(JT)- 1 0~
the former case. For this purpose we assume that a total of nint integration points is
adopted for the entire body (this notation is used here to distinguish nint from n, the
[-oux oux
- number of nodal points). With (20.8}, the stiffness matrix K may be written as
oy a, n inl

Use of (20.23) and (20.24) then gives K = L aiBJDiBi (20.28)

i= 1
I
oux = - 2 c,; OUx = - 2 c~ ( where Bi and D; indicate that B and D are evaluated at Gauss point i. Moreover,
ox x2 - x1 oy Y3 - Yt the parameters cxi depend on the Gauss weight and det J; which are both positive, i.e.
and as uY = 0, we find that CX; >0 (20.29)

2 From (20.28), the elastic strain energy U given by (20.18) becomes

Yxy =- C~ (20.25) n int nint
Y3- Y1
U =fa TKa = tL cxia TB[D;B;a =t L a;e[Diei (20.30)
I.e. i= 1 i= 1

ll~ = q = O =0 (20.26) where

(20.31)
This result may, in fact, be concluded directly by investigating Figure 20.6(a). Use
of (20.26) in (20.21) yields
By definition, zero-energy modes exist when the strain ~nerg~ U =.?
for a ": 0
K•a• = 0; i.e. U = ia•TK•a• = 0 (20.27) and we want to identify such modes. As the constitutive matnx D IS pos1t1ve defimte
(cf. (13.11)), it follows from (20.29) and (20.30) that U = 0 if, an~ only if, ll; = 0 ~or
It appears that the displacements given by (20.22) and illustrated in Figure 20.6(a) i = 1, ... , nint· According to (20.31), this requires a non-trivial solutiOn of the followmg
result in a spurious zero-energy mode. Another such mode is illustrated in Figure homogeneous system of equations:
20.6(b).
The existence of spurious zero-energy modes should, of course, be avoided in an
FE model, and we also observe that these zero-energy modes emerge as a result of
reduced integration. Moreover, as the body provides no resistance at all to spurious (20.32)
zero-energy modes, the effect of reduced integration is to soften the stiffness of the
FE model. However, we have previously shown that with an exact integration of the

l
400 Introduction to the finite element method Numerical integration 401
In the following let us consider the analysis of two-dimensional elasticity. The Table 20.2 Spurious modes for the structures in Figure 20.7
results obtained are, however, easily carried over to one- and three-dimensional
elasticity as well as to heat flow. For two-dimensional elasticity, and with n denoting Spurious modes Spurious modes
the number of nodal points for the entire body, the dimensions ofB; and a are given by for four-node element for eight-node element

B; = 3 x 2n; a = 2n x 1 (20.33) Structure (a) 2

Structure (b) 3 none
(cf. ( 15.13) and (15.14 )). It appears that the dimensions of the matrices entering (20.32) Structure (c) none none
arc given as shown below:
Bl
Bz Table 20.3 Gauss integration techniques used for two-dimensional elements
a=O (20.34)
Reduced integration used
Reliable in practice (with spurious
Element integration order zero-energy mode (s))
(3n;01 X 2n)(2n X 1) = 3n;01 X 1
That is, in this homogeneous system of equations, we have 2n unknowns and 3n; 01
equations. It is evident that if2n > 3n;"" then there are more unknowns than equations
and we have at least 2n- 3n;01 non-trivial solutions of a (cf. (2.65)). Therefore, we have
Four-node
I>----I 2 X 2

at least 2n - 3n;01 zero-energy modes (20.35) Four-node 2x2

I distorted
In the case of a four-node isoparametric element with one point integration, we
haven= 4 and n; 01 = 1, i.e. we have at least 2 x 4- 3 x 1 = 5 zero-energy modes. (
Eight-node 3 X 3 2 X 2

H a)

H X

Eight-node
distorted
3 X 3 2 X 2

1I 1 b)
X X X X
Nine-node 3 X 3 2 X 2
X X
X X )( X

Nine-node 3 X 3 2 X 2
distorted

c)
Of these, we have already identified the three modes corresponding to rigid-body
motions (cf. Figure 20.5 ), as well as the two spurious zero-energy modes (cf. Figure
20.6).
Figure 20.7 Illustration of identification of spurious zero-energy modes In practice, the structure is supported at some nodal points, i.e. some of the
 (
402 Intr o duction to the finite element method

components of the a-vector are prescribed. Let npre denote the number of prescribed
displacements. Then we conclude from (20.35) that we have

at least (2n- npre) - 3nint zero-energy modes (20.36)

In practice, the number npre is sufficiently large to prevent rigid-body motions (i.e.
npre ;?: 3) and the zero-energy modes given by (20.36) are therefore spurious modes.
I References
I
We recall that (20.36) refers to two-dimensional solid mechanics and that similar
expressions may be derived for other situations.
In order to illustrate the use of (20.36), the structures shown in Figure 20.7 are
investigated. All these structures are supported so that rigid-body motions are
prevented, i.e. (20.36) provides the number of spurious zero-energy modes. As an
example, consider the structure modelled by four-node elements with one point Adini, A. and Clough, R. W. ( 1961) 'Analysis of plate bending by the finite element method',
integration shown in Figure 20.7(b). We haven= 6· n = 3· n. = 2· that is (20 36) Report to National Science Foundation, USA, G.7337.
' pre ' 1nt ' ' •
gives (2 x 6 - 3) - 3 x 2 = 3 spurious zero-energy modes. If we proceed in this Barsoum, R. S. (1976) 'On the use of isoparal!letric finite elements in linear fracture mechanics',
manner with all the structures shown in Figure 20.7, the results shown in Table 20.2 International Journal for Numerical Methods in Engineering, 10, 25- 37.
are obtained. Bathe, K.-J. ( 1982) Finite Element Procedures in Engineering Analysis, Prentice Hall: Englewood
It appears that expression (20.36) may be used to determine quickly whether Cliffs, NJ.
spurious zero-energy modes are present, and in this case reduced integration is Bear, J. ( 1979) Hydraulics of Groundwater, McGraw-Hill: New York.
evidently to be avoided. Becker, E. B., Carey, G. F. and Oden, J. T. ( 1981) Finite Elements. An Introduction, vol. 1,
Prentice Hall: Englewood Cliffs, NJ.
Boresi, A. P., Sidebottom, 0. M., Seely, F. B. and Smith, J. 0. ( 1978) Advanced Mechanics of
20.4.1 Suggested order of Gauss integration of Materials, 3rd edn, Wiley: New York.
Carslaw, H. S. and Jaeger, J. C. (1959) Conduction of Heat in Solids, 2nd edn, Clarendon Press:
isoparametric elements
Oxford.
Cook, R. D., Malkus, D. S. and Plesha, M. E. (1989) Concepts and Applications of Finite
From the discussion above, we may conclude that full integration is always a reliable Element Analysis, 3rd edn, Wiley: New York. .
technique by which pitfalls are avoided. However, reduced integration may improve Crandall, S. H. (1956) Engineering Analysis. A Survey of Numerical Procedures, McGraw-Htll:
the accuracy, provided that spurious zero-energy modes are not present. We also New York.
observe that it is advantageous to distort the isoparametric elements as little as Crisfield, M. A. (1986) Finite Elements and Solutions Procedures for Structural Analysis, vol. 1,
possible. Table 20.3, taken from Bathe (1982, p. 286), summarizes some of these Linear Analysis, Pineridge Press. .
general conclusions. Dahlquist, G . and Bjorck, A. ( 1974) Numerical Analysis, Prentice Hal~: Englewoo~ Chffs, ~J.
In conclusion, the discussion above suggests that choice of the order of integration Ergatoudis, J. G., Irons, B. M. and Zienkiewicz, 0. C. ( 1968) ' Curved tsoparametnc q uadnlateral
elements for finite element analysis', International Journal ofSolids and Structures, 4, 31-42.
is far from being trivial and, in effect, many special techniques have been developed
Finlayson, B. A. and Scriven, L. E. ( 1966) 'The method of weighted residuals - a review ',
for efficient reduced integration techniques. Such issues are discussed at length by
Applied Mechanics Reviews, 19, 735-48. .
Zienkiewicz and Taylor (1989), Hughes (1987) as well as by Bathe (1982). Finlayson, B. A. (1972) The Method of Weighted Residuals and Variational Principles, Academtc
Press: New York.
Froberg, C.-F. (1965) Introduction to Numerical Analysis, Addison-Wesley: Readi.ng, MA.
Fung, Y. C. (1965) Foundations of Solid Mechanics, Prentice Hall :. Engle.~o~d Chffs, NJ.
Galerkin, B. G. (1915) 'Series solution of some· problems of elasttc equthbnum of rods and
plates' (in Russian), Vestn. Inzh. Tech., 19, 897- 908. .
Gallagher, R. H. (1975) Finite Element Analysis. Fundamentals, Prentice Hall: Englewood Chffs,
NJ.
Henshell, R. D. and Shaw, K. G. (1975) 'Crack tip elements are unnecessary', International
Journal for Numerical Methods in Engineering', 9, 495-507.

403

i
404 References References 405

Hildebrand, F. B. (1965) Methods of Applied Mathematics, 2nd edn, Prentice Hall: Englewood Timoshenko, S. P. and Gere, J. M. ( 1972) Mechanics ofMaterials, Van Nostrand: New York.
Cliffs, NJ. Timoshenko, S. P. and Woinowsky-Krieger, S. (1959) Theory of Plates and Shells, 2nd edn,
Hughes, T. 1. R. (1987) The Finite Element Method. Linear Static and Dynamic Finite Element McGraw-Hill: New York.
Analysis, Prentice Hall: Englewood Cliffs, NJ. Turner, M. J., Clough, R. W., Martin, H. C. and Topp, L. P . (1956) ' Stiffness and deflection
Hughes, W. F. and Brighton, J. A. ( 1967) Theory and Problems of Fluid Dynamics, Schaum. analysis of complex structures', Journal of Aeronautical Sciences, 23, 805- 23.
Irons, B. M. (1966a) 'Engineering application of numerical integration in stiffness method', Zienkiewicz, 0. C. ( 1970) 'The finite element method: From intuition to generality', Applied
Journal of AIAA, 14, 2035- 7. Me chanics Review, 23, 249- 56.
Irons, B. M. (1966b) 'Numerical integration applied to finite element methods', Conference on Zienkiewicz, 0. C. (1977) The Finite Element Method, 3rd edn, McGraw-Hill: New York.
the Use of Digital Computers in Structural Engineering, University of Newcastle. Zienkiewicz, 0. C. ( 1983) 'The generalized finite element method - State of art and future
Johnson, C. ( 1987) Numerical Solutions of Partial Differential Equations by the Finite Element directions', Journal of Applied Mechanics, 50, 1210-17.
Method, Student litteratur: Lund. Zienkiewicz, 0. C. and Taylor, R. L. (1989) The Finite Element Method, 4th edn, vol. 1,
Kaplan, W. (1981) Advanced Mathematics for Engineers, Addison-Wesley: Reading, MA. McGraw-Hill: New York.
Kirchhoff, G. (1850) 'Ober das Gleichgwicht und die Bewegung einer elastichen Scheibe',
Journal fur die reine und angewandte Mathematik (Grelle), 40, 51-8.
Kollbrunner, C. F. and Hajdin, N. (1972) Diinnwangidge stabe, Springer Verlag: Berlin.
Kreyszig, E. (1979) Advanced Engineering Analysis, 4th edn, Wiley: New York.
Lekhnitskii, S. G. (1981) Theory of Elasticity of an Anisotropic Body, Mir: Moscow.
Love, A. E. H. (1944) A Treatise on the Mathematical Theory of Elasticity, 4th edn, Dover:
New York.
Malvern, L. E. (1969) Introduction to the Mechanics of a Continuous Medium, Prentice Hall:
Englewood Cliffs, NJ.
Melosh, R. J. (1963) 'Basis of derivation of matrices for the direct stiffness method', Journal
of AIAA, 1, 1631 - 7.
Mindlin, R. D . (1951) 'Influence of rotary inertia and shear on flexural motion of isotropic
elastic plates', Journal of Applied Mechanics, 18, 31-8.
Morley, L. S. D. (1971) 'The constant bending-moment plate bending element', Journal of
Strain Analysis, 6, no. 1.
Owen, D. R. J. and Hinton, E. (1980) Finite Elements in Plasticity, Pineridge Press. I
Press, W. H., Flannery, B. P., Teukolsky, S. A. and Vetterling, W. T. (1986) Numerical Recipes,
Cambridge: Cambridge University Press.
Przemieniecki, J. S. (1968) Theory of Matrix Structural Analysis, McGraw-Hill: New York.
Reissner, E. ( 1945 ) 'The effect of transverse shear deformations on the bending of elastic plates',
Journal of Applied Mechanics, 12, A69- 77.
Roark, R. 1. (1975) Formulas for Stress and Strain, 5th edn, McGraw-Hill: New York.
Segerlind, L. J. (1976) Applied Finite Element Analysis, Wiley: New York.
Sokolnikoff, I. S. and Redheffer, R. M. (1958) Mathematics and Physics of Modern Engineering,
McGraw-Hill: New York.
Sokolnikoff, T. S. (1946) Mathematical Theory of Elasticity, McGraw-Hill: New York.
Spencer, A. J. M. (1980) Continuum Mechanics, Longman: Harlow. ·
Stasa, F. L. ( 1985) Applied Finite Element Analysisfor Engineers, CBS International Editions.
Strang, G. (1980) Linear Algebra and its Applications, 2nd edn, Academic Press: New York.
Strang, G. and Fix, G. J. (1973) An Analysis of the Finite Element Method, Prentice Hall:
Englewood Cliffs, NJ.
Taig, I. C. (1961) 'Structural analysis by the matrix displacement method', English Electric
Aviation Report no. So17.
Thelandersson, S. ( 1984) Konstruktionsberiikningar med dator, Studentlitteratur: Lund.
Timoshenko, S. and Goodier, J..N. (1970) Theory of Elasticity, 3rd edn, McGraw-Hill: New
York.

i
I Index
I

addition of matrices, 13 C0 -continuity, 202, 213, 306

adjoint matrix, 19 C'-continuity, 323, 360
anisotropy, 79, 251 C"-continuity, 202
arc length, 72 C-matrix method, 100, 108
aspect ratio, 224 calculus of variation, 204
assembling, 33, 182 Cauchy elasticity, 249
axially loaded bar, 52, 193 Cauchy's formula, 240
axisymmetry, 299, 305 centre of twist, 262
cofactor, 17
back-substitution, 21 column, 11
balance principle, 176, 210 column matrix, ll
bandwidth, 225, 231 column operation, 18
basis functions, 144 compatibility, 30, 93, 283, 324, '160, 375
beam element, 324 completeness, 92, 283, 323, 360, 376
beam theory complete polynomial, 97
Bernoulli, 314, 317 compliance matrix, 250
Mindlin-Reissner, 318 computational cost, 229
Timoshenko, 318 condensation, 227
bending moment, 312, 336 conforming, 93, 283, 324, 361, 375
bending stiffness, 316 conservation equation, 49
Bernoulli's assumption, 314, 339 consistent load vector, 178
Bernoulli beam theory, 317 constant strain element, 288
biharmonic equation, 344 constitutive matrix
bilinear, 126 diffusion, 87
body force, 235, 239 elasticity, 249
boundary conditions electric current, 87
Dirichlet, 59, 86 flow in porous media, 89
essential, 59, 86, 297, 321, 358 heat, 78
for beams, 321 continuity, 202
for plates, 347, 352, 358 continuous system, 4
for stresses, 240 contour curve, 67
for torsion, 266 convection, 83, 198, 219, 222
Kirchhoff, 352 convection coefficient, 198
mixed, 59, 86 convergence, 91, 93, 282, 323, 360, 374
natural, 59, 86, 297, 319, 352 coordinates
boundary integral, 71 global, 37, 38, 364
boundary value problems, I local, 37, 38, 191, 364
boundary vector, 160, 208, 274, 297, 304, 321, crack, 373
359 cross product, 3 77
brick elements, 137 cubic element, 115

407

J
408 Index In dex 409
curvature, 317 flexible integration (continued) subtraction, 13
curvature matrix, 341 membrane, 88, 270 Gauss, 391 symmetric, 12
string, 54, 63 of matrix, 25 transformation, 40
Darcy's law, 88, 89 flow diagram, 197 Newton-Cotes, 389 transposed, 11
degrees of freedom, 4 fluid flow numerical, 384, 386, 387 two-dimensional, I I
determinant, 16 in porous media, 87, 88 order of, 394, 402 unit, 12
diagonal matrix, 12 incompressible, 88, 89 point, 387 zero, 12
differentiation of matrix, 25 flux, 50, 77 reduced, 396 Melosh element, 126, 289
diffusion, 51, 87, 88 flux vector, 76 transformation of, 376, 380, 384 membrane
Dirac's delta function, 147, 177, 210, 332 force vector, 32, 161, 175,208,274,298, 305,321,359 action, 342
Dirichlet boundary conditions, 59, 86 four-node element, I 26, 137, 289, 366 Jacobian, 365 analogy, 269
discontinuity, 59 Fourier's law, 50, 63, 77, 88 Jacobian matrix, 365 flexible, 88, 270
discrete system, 4 full integration, 396 mesh, 1, 223
displacement, 28, 53 kinematic Mindlin- Reissner
components, 243, 282 Galerkin's method, 143, 152, 159, 208, 273, 296, boundary conditions, 297, 304, 321, 348, 358 beam theory, 318
gradient, 243 302, 320, 357 relations, 53, 24 7 plate theory, 344
vector, 243, 282 Gauss Kirchhoff mixed boundary conditions, 59, 86
displacement method, 35 elimination, 20, 226 boundary conditions, 352 Mohr's circle of stress, 345
divergence, 73, 74 integration, 391 plate theory, 344 moment of inertia, 316
divergence theorem of Gauss, 73, 75 points, 392
domain, 364 general purpose program, 6 Lagrange natural boundary conditions, 59, 86, 295, 319, 352
global element, 117, 126, 131, 132 Navier's equations, 260
effective shear force, 352 coordinates, 37, 38 interpolation, 116, 389 Neumann boundary conditions, 59, 86
eight-node element, 134, 137, 370, 374 domain, 364 Laplace Newton's convection boundary condition, 83,
elasticity nodal numbering, 187 equation, 82, 84 198, 219
coefficients, 249 shape functions, 103, 104, 111, 112, 140, 286 expansion formulas, I 7 Newton-Cotes integration, 389
linear, 248 shape function matrix, 104, 113, 140, 286 least square method, 151 nine-node element, 131
three-dimensional, 292, 295 stiffness matrix, 33, 41, 160 length of vector, 14 nodal displacement vector, 29
two-dimensional, 299, 302 gradient, 66, 70 line source, 210 normal
elastic bar, 37, 52, 193 Green elasticity, 249 linear strains, 245
elastic strain energy, 249, 396 Green- Gauss theorem, 74, 75 elastic fracture mechanics, 373 stresses, 237
electric currents, 51, 63, 87, 88 elasticity, 248 numerical integration, 384, 386, 387
element heat elements, 58, 118, 137, 288
boundary vector, 186, 212, 275, 299, 306, 322 flow, 48, 76, 157, 206 load vector, 160, 194, 208, 274, 297, 304, 321, 359 Ohm's law, 51, 63, 87, 88
force vector, 186, 212, 275, 299, 306, 322 flux, 50, 77 local coordinates, 38, 191, 369 one-dimensional elements, 98
load vector, 186, 212, 275, 299, 306, 322 flux vector, 76 local nodal point numbering, 36, 187 operation
nodal displacement vector, 29, 285 Hermite interpolation, 328 local stiffness matrix, 39 column, 18
shape functions, 99, 138, 285 homogeneous equation system, 20, 23 row, 18
shape function matrix, 100, 138, 285 Hooke's law, 28, 37, 53, 63, 249 mapping, 364 operator, 143
stiffness matrix, 29, 186, 212, 275, 299, 306, 322 hydraulic potential, 88 matrix, 11 order of integration, 394, 402
elimination Gauss, 20, 226 hyperelasticity, 249 addition, 13 orthogonal matrix, 19
equation system, 19, 23 adjoint of, 19 orthogonality, 145
equilibrium inhomogeneous equation system, 20 cofactor of, 17 orthotropy, 79, 252
conditions, 241, 242, 312, 335 initial column, 11 out-of-plane strain, 255
weak form of, 292, 299, 318, 349 strains, 253 diagonal, 12
essential boundary conditions, 59, 86, 297, 321, strain vector, 298, 304 differential operator, 241, 242, 246, 247, 343 parasitic terms, 97
358 value problems, 1 differentiation of, 25 parent domain, 364
expanded element initialization, 198 dimension of, 11 partitioning, 24, 171, 226
boundary vector, 180 in-plane strains, 255, 256 integration of, 25 Pascal's triangle, 97
FE-formulation, 31, 179 in-plane stresses, 255, 256 inverse, 18 patch test, 94, 283
force vector, 31, 180, 182 isoparametric elements, 136, 141, 364, 374 minor of, 17 piezometer, 87
load vector, 180, 211 isotropy, 80, 252 multiplication, I 3 piezometric head, 87
stiffness matrix, 31, 180, 21 1 integral one-dimensional, 11 pivot elements, 21
expansion formulas of Laplace, 17 area, 70 orthogonal, 19 plane
boundary, 71, 73 partitioning, 24 strain, 247, 255, 299
Fick's law, 51, 63, 87, 88 volume, 75 positive definite, 23 stress, 241, 254, 299
field equations, 257 integration positive semi-definite, 24 plate theory
finite element mesh, 1, 223 by parts, 56, 65 singular, 19 Kirchhoff, 344
flexibility matrix, 250 full, 396 square, 12 Mindlin-Reissner, 344
410 Index

point collocation, 147 strength of source, 177, 210

point force, 298, 305, 332 stress
point source, 177 components, 237
Poisson's function, 264
equation, 82, 84, 265 tensor, 235, 237
ratio, 252 string, 54, 63
positive strong form, 51, 83, 271
definite, 23 subdomain collocation, 149
semi-definite, 24 submatrix, 24
post-processor, 198 subtrac.tion of matrices, 13
Prandtl's surface forces, 240, 243
soap film analogy, 269 symmetric matrix, 12
stress function, 264 symmetry plane, 251
pre-processor, 198
principle thermal conductivity, 50, 80
moments, 346 thermal expansion coefficient, 254
stresses, 346 thermal strains, 253, 258, 303
thermo-elasticity, 254
quadratic form, 23 three-dimensional elements, 136, 286, 373
quadratic one-dimensional element, 106 three-node element, 118, 288
quadrature, 387 Timoshenko beam theory, 318, 327
quartic one-dimensional elements, 117 torque, 267
quasi-harmonic equation, 82, 84 torsion, 261
torsional rigidity, 268
rate of convergence, 94, 283 traction vector, 235, 240
rate of twist, 264 transformation of integrals, 376, 380
Rayleigh- Ritz, 204 transformation matrix, 40
reduced integration, 396 transpose, 11
remainder, 388 transverse isotropy, 252
residual, 145 trial functions, 144
rigid-body motions, 34, 195, 246, 283,298, 323, 360 triangularization, 21, 227
row truss analysis, 37
matrix, 12 twisting moment, 336
operation, 18 two-dimensional elements, 118, 284, 366, 370

Saint-Venant torsion, 261 unit matrix, 12

scalar field, 65
scalar product, 14 variational calculus, 204
semi-inverse method, 271 vector length, 14
serendipity element, 135, 370 velocity potential, 89
shear virtual work equation, 295, 302
force, 312, 336 Vlasov theory, 271
modulus, 253
strains, 246 weak form
stresses, 237 beam theory, 319
shell element, 342 one-dimensional heat, 57
singular matrix, 19 plate theory, 351
skyline, 229 three-dimensional elasticity, 295
soap film analogy, 269 three-dimensional heat, 86
solid mechanics, 256, 259 two-dimensional elasticity, 302
springs, 27 two-dimensional heat, 85
spurious zero-energy modes, 395, 396 weight, 388
static weight function , 56, 85, 143
boundary conditions, 240, 297,304,321, 348,358 weighted residual method, 142
condensation, 227
stiffness matrix, 29, 31, 33, 160, 170, 208, 274, 297, Young's modulus, 53, 252
304, 321, 359, 384
strains, 243 zero matrix, 12
strain energy, 249, 396 zero-energy modes, 396