1

Preliminaries
A journey of a thousand miles
begins with a single step
Lao-tzu, Chinese philosopher

1.1 INTRODUCTION
Nearly all physical phenomena occurring in nature can be described by differential
equations and boundary conditions. In the solution of these boundary value problems
we aim to determine a response to given boundary conditions. For example we may be
interested in determining the response of the rock mass due to the excavation of a tunnel,
or the response of a structure to dynamic excitations of its foundations (caused by an
earthquake). Analytical solutions of boundary value problems, i.e. solutions that satisfy
both the differential equations (DE) and the boundary conditions (BCs), can only be
obtained for few problems with very simple boundary conditions. For example,
analytical solutions exist for the excavation of a circular tunnel in a homogeneous rock
mass, not really a realistic scenario for practical tunnelling. To be able to solve real life
problems, the engineer must revert to approximate solutions. Two approaches can be
taken: instead of satisfying both the DE and the BCs, one can attempt to satisfy only one
of the two and minimise the error in satisfying the other one. In the first approach (based
on the original idea of Ritz1) solutions are proposed that satisfy the boundary conditions
exactly. The error in satisfying the differential equation is then minimised. This is the
well known Finite Element Method. In the alternative (proposed by Trefftz2), the
assumed functions satisfy the DE exactly and the error in the satisfaction of the
boundary conditions is minimised.
Most readers of this book will be familiar with the finite element method. In the most
common version of this method we subdivide the domain into elements and approximate
2 The Boundary Element Method with Programming

the response to a specified loading with functions which are defined at element level,
i.e., are piecewise continuous. This subdivision is necessary because in practice it is
impossible to determine functions that cover the whole domain and at the same time
satisfy the boundary conditions (as originally proposed by Ritz). The parameters of these
functions, which are the values of the unknowns at the nodes where elements are
connected to each other, are determined by minimising the error in satisfying the DE.
This can be done using residual methods, where the integral of the error is minimised
and this involves a domain integral. A violation of the DE may occur at any point in the
domain, but the variation of the unknown is chosen in such a way that the error in the
satisfaction of the DE over the whole domain is a minimum. In continuum mechanics,
for example, this means that the chosen functions will usually not satisfy exactly the
equilibrium conditions at specified points.
Figure 1.1 shows an example of a finite element mesh for the three-dimensional
analysis of sequential excavation and construction of a tunnel. A plane of symmetry is
applied, so that only half of the tunnel is discretised. Note that to model the rock mass
through which the tunnel is driven, which for all practical purposes can be assumed to be
infinite, we must make a 'box' of solid elements. At the outer boundaries of this box,
unless we use infinite elements, we either set displacements to zero or apply stress
boundary conditions, which represent the in situ stress. The mesh shown here has
approximately 100 000 degrees of freedom and a solution took several hours on a PC.
Note that small jumps occur in the contours of maximum compressive stress, between
elements indicating a lack of satisfaction of equilibrium locally.
The second approach to solving this problem (based on the original idea of Trefftz)
does not require the subdivision of the domain into elements because the functions used
for approximating the solution inside the domain are chosen to be those which exactly
satisfy the governing differential equations. In a similar way as with the FEM the error
in satisfying the boundary conditions is minimised and this now involves a boundary
integral. Numerically, this integral can be evaluated by subdividing the boundary into
elements over which the values (for example, tractions or displacements in the case of
continuum mechanics) are interpolated, much in the same way as with the FEM. The
advantage of the method is obvious: the dimensionality of the problem is reduced by one
order, i.e. only a surface instead of a volume integral is required. This means that the
number of unknowns is reduced dramatically, especially for three-dimensional
problems, because unknowns occur only on the problem boundary. Other advantages are
that the DE is satisfied exactly everywhere in the domain and that infinite domain
problems are easy to deal with.
As an example, Figure 1.2 shows the boundary element mesh for the same tunnel as
analysed by the FEM in Figure 1.1. This mesh has approx. 1000 degrees of freedom and
took 3 minutes to solve on a PC. The stress contours computed and drawn on the
excavation surface and a user defined plane inside the rock mass show no jumps as they
are seen in FEM results. Since functions must be found which exactly satisfy the
governing differential equation (DE) the BEM requires a solution of the DE. This
solution must be as simple as possible because, as will be seen in the chapter on
implementation, this is crucial for efficiency. Unfortunately, the simplest solutions
which we can find (fundamental solutions) are due to concentrated loads or sources and
PRELIMINARIES 3

are singular, i.e., have infinite values at certain points. This property has to be taken into
account when integrating these functions over boundary elements. This will make the
numerical integration procedure more complicated than is the case with finite elements.

Figure 1.1 Finite element mesh for the analysis of tunnel excavation. Left side: mesh
with contours of z-displacement; right side: detail with contours of
maximum compressive stress

Figure 1.2 Boundary element mesh for the simulation of tunnel excavation with
contours of maximum compressive stress plotted on excavation surface
and result planes
4 The Boundary Element Method with Programming

There has been a general misconception that because a fundamental solution of the
problem must exist for the BEM to work, the method can only be applied to linear
problems with homogeneous material. As will be shown in this book, non-linear
problems can almost as easily be solved as with the FEM, by the repeated solution of
linear problems and special methods may be employed to solve problems with
heterogeneous material properties.

1.2 OVERVIEW OF BOOK

This book is designed to be used as basis for a course on the BEM or for self study.
It is recommended that chapters be read consecutively as later chapters build on material
discussed earlier. Throughout the book, the reader will build a suite of subprograms,
which perform the various tasks needed for the numerical implementation of the BEM.
Various exercises are included which allow the reader to test the programs written and
experience how the method works.
We start with an introduction to the FORTRAN 95 programming language.
FORTRAN, which stands for FORMula TRANslation is still the most widely used
language for programming engineering applications and is easier to learn and more
efficient than other high level languages such as C++. However, there is no reason why
the procedures outlined in some detail in this book could not be implemented in another
language.
The next chapter deals with the way in which we can describe the geometrical
boundary of a problem and boundary conditions in a numerical way. This is done by
subdividing the surface into small elements and by interpolating between nodal values.
This is essential for the later treatment of integral equations. With the aid of the
examples we can not only test the subroutines developed but also get an understanding
of the error introduced by the approximations used to describe boundaries.
Another fundamental building block is the description of the material response. In
Chapter 4 we introduce basic concepts of elasticity and potential flow and develop
fundamental solutions, that is, simple solutions which satisfy the governing differential
equations. These will be central to our subsequent deliberations.
Next we introduce the concepts of boundary element methods using the method
originally proposed by Trefftz. Although this very simple method cannot be used for
general purpose programs, it serves very well to explain the fundamental ideas of the
method. A small computer program can be developed to solve some simple problems.
Again, this will serve as a tool for learning by experience.
The direct boundary element method used in the majority of BEM software is
introduced next. Here we will use the reciprocal theorem by Betti, which is well known
to engineers to obtain an integral equation. The major task in the implementation
however, is to solve the integral equations numerically.
The next chapter on numerical implementation therefore deals with the evaluation of
integrals using numerical integration. Those familiar with isoparametric finite elements
will recognise the Guass Quadrature method used. However, in contrast to its use in the
PRELIMINARIES 5

FEM, one must be very careful to select the number of integration points, as they are
dependent on how close the singularity is to the integration region. This is the most
difficult and crucial part in the implementation of the BEM. The integration over the
boundary surface is carried out over a boundary element and the contributions of all
elements which describe a boundary are then added. We will see that this is very similar
to the assembly procedure in the FEM.
After the numerical treatment of the integral equations we end up with a system of
equations. In contrast to the FEM, the coefficient matrix is fully populated and
unsymmetrical. Standard Gauss elimination can be used but, for large systems, the
storage requirement and the computation times may be reduced considerably by iterative
solvers, such as conjugate gradient methods. Such special solution techniques are
introduced in the next chapter. Here we also find that the method is embarrassingly
parallelisable i.e. that excellent speed up rates can be achieved with special hardware.
The primary results obtained from the analysis are values of displacement or traction
at the boundary depending on the boundary condition specified. In contrast to the FEM,
primary results do not include values in the interior of the domain but these are
computed by post-processing. In Chapter 9 it is explained how the stresses at the
boundary and in the interior can be obtained from boundary displacements and tractions.
This is indeed an advantage of the method, because the user has free choice of the
locations where results are obtained.
We now have all the building blocks together and are able to compile a computer
program that is able to solve two and three-dimensional problems in elasticity and
potential flow, depending on which fundamental solution is used. In Chapter 10 we
apply the program developed to test examples and find out what level of accuracy can be
obtained in comparison with the FEM.
For inhomogeneous domains, where we can not obtain a fundamental solution, we
introduce the concept of multiple regions, where the domain is subdivided into sub-
regions, similar to the FEM. There is an additional advantage in this concept, because
sparseness is introduced in the system of equations. We will also find out in a later
chapter that the multi-region method allows contact and excavation problems to be
solved in an elegant way.
In the next chapter we deal with problems that involve corners and geometry which
changes with time, as is the application to sequential excavation/construction of a tunnel.
Because elements only exist on the boundary the BEM has difficulty dealing with
problems where forces are applied inside the domain. These forces can be loosely
termed body forces. It will be shown that an additional volume integral has to be
considered. For body forces that are constant the volume integral can be transformed
into a surface integral. However, if the body forces are not constant throughout the
domain the volume integral needs to be evaluated numerically. This can be done by
using internal cells, which look like finite elements, but do not involve any additional
degrees of freedom, as they are only used for integration. The implementation of this
procedure, discussed in chapter 13 also allows the solution of problems in elasto- and
visco-plasticity. Body forces of a different kind (mass forces) occur in the case of
dynamics, but their treatment with the BEM is quite different to the FEM and this is
discussed in Chapter 14.
6 The Boundary Element Method with Programming

In Chapter 15 we show that the solution of non-linear problems follows similar

procedures as in the FEM and that the general solution algorithm is similar. Here two
types of non-linear problems are discussed in more detail: plasticity and contact
problems.
It is possible to couple the BEM with the FEM thus getting the best of both worlds.
In Chapter 16, methods of coupling are presented. Basically, a stiffness matrix of the BE
region is obtained and assembled with the FEM stiffness matrices. Since many general
purpose programs allow the input of a user defined element stiffness matrix this may be
used to extend the capabilities of a readers FEM code.
To demonstrate that the method also works for large scale industrial problems,
Chapter 17 shows some applications of the boundary and coupled method in
engineering. The purpose of this chapter is twofold: firstly it shows how complex
problems, as they invariably occur in real life, can be simplified and how a suitable
boundary element mesh is obtained. Secondly it shows the advantage of the BEM and
the coupled BEM/FEM in terms of user friendliness and computing time.
The last chapter deals with topics which were still subject to research at the writing
of the book. The first deals with the efficient treatment of heterogeneous ground
conditions the other with the consideration of linear inclusions such as reinforcement
and rock bolts. The application in piezo-electricity shows the flexibility of the method to
deal with any problem whose fundamental solution is known.
By the end of this book the reader should have an understanding of how the method
works, of its potential and how it can be implemented into a computer program.

1.3 MATHEMATICAL PRELIMINARIES

A good consistent notation is essential to any textbook. For the development and
explanation of numerical methods two notations are used by engineers: matrix and
tensor notation. Traditionally, textbooks on the BEM have use tensor notation, whereas
those about the FEM have used matrices, although this is rapidly changing. The main
notation chosen for this book is the matrix notation.
There are two reasons for this: firstly, the book which is probably still the most
widely read on numerical modelling, The Finite Element Method, by O.C.
Zienkiewicz and R.L Taylor3, uses matrix notation throughout. Since we hope to attract
more engineers to the BEM, this was one motivation. The other reason is that books on
the BEM that use tensor notation have to revert to matrix notation at some stage, for
example when discussing the assembly of the system of equations. Thus the book
attempts to avoid two different notations.
However, when discussing fundamental solutions and their derivatives it transpires
that tensor notation is much easier to use. Therefore in this book we have made a
compromise in that for this case only we revert to a simplified version of the tensor
notation.
In the following we discuss some basic mathematics which will be used in this book
and also attempt a comparison of matrix and tensor notation.
PRELIMINARIES 7

1.3.1 Vector algebra

Vectors are used to represent a displacement/force or to define the position of a point

relative to a set of Cartesian axes. We define the position of a point in 3-D space with
respect to Cartesian axes x, y, z (Figure 1.3) as

x
(2.1)
x y

z

i3
x
z

i2
y

x
i1

Figure 1.3 Position vector x defining a point in space

Alternatively, we may represent the point in terms of Cartesian coordinates xi, where
i=1,2,3 (the last number is also referred to as range).
The components are specified with respect to a set of orthogonal coordinate axes,
which are defined by base vectors of unit length, ii and which have the property:

1 for i j
i i x i j G ij (2.2)
0 for i z j
where   x   denotes the scalar (dot) product

ii x i j i1x i2 x  i1 y i2 y  i1z i2 z (2.3)

and Gij is known as the Kronecker delta.

Vector x may then be represented in indicial notation as

3
x x i
i 1
i i xi i i (2.4)
8 The Boundary Element Method with Programming

where the Einstein summation convention has been used for the last term. This
convention specifies that for any index which is repeated and which does not appear on
the left hand side, a summation of all terms within the range is implied.
Another vector quantity is the displacement which can be written either as

u x
(2.5)
u u y
u
z

in matrix notation or u ui i i in indicial notation

Coordinate transformation

If we want to express the location of a point, x in another orthogonal coordinate system

( x ) the directions of which are given by unit vectors v1 , v2 , v3 then in matrix notation
we write

x Tg x (2.6)

where the transformation matrix is defined as

Tg > v1 v 2 v 3 @ (2.7)

Alternatively, we may write in indicial notation

xi / ij x j (2.8)

where
/ ij vi x i j (2.9)

Projection of one vector onto another

If we want to compute the projection of a vector onto a direction specified by a unit

vector v, then it is very convenient to use the dot product. For example, the component,
u c of the displacement u in the direction specified by v is given by

uc u x v (2.10)

The angle T between the two vectors is computed by

1
cosT ux v (2.11)
u
PRELIMINARIES 9

where the length of vector u is given by

2 2 2 (2.12)
u ux  u y  uz

v
T

uc

Figure 1.4 Projection of vector

Derivatives of vectors

The derivatives of the displacement vector may be written as

wu x wu x wu x
wy
wx wz
wu wu y wu wu y wu wu y
u, x ; u, y ; u, z (2.13)
wx wx wy wy wz wz
wu z wu z wu z
wx wz
wy

In indicial notation we simply write

wui
ui , j (2.14)
wx j
10 The Boundary Element Method with Programming

1.3.2 Stress and strain

Stresses and strains are tensorial quantities. In the indicial notation the strain tensor is
defined by
1
H ij ui , j  u j ,i  (2.15)
2
In this book, however, we use a notation originally proposed by Timoshenko4.
We define a pseudo-vector of strain, i.e., a matrix with one column:

wu x
wx
wu
y
Hx
H wy
y wu z
Hz
wz
H wu wu (2.16)
J xy x y

J yz wy wx
wu y wu
J xz  z
wz wy
wu x wu z
wz  wx

Note that in the pseudo-vector notation we only have 6 strain components, whereas
the symmetric strain tensor has 9. Also note that the term is missing for the shear
strains in order to achieve consistency between the tensor and matrix operations. The
index number of the location of the strain or stress components for matrix notation and
tensor notation is given in Table 1.1

Table 1.1 Index numbering for strain and stress

Notation Index number

Matrix 1 2 3 4 5 6
Tensor 11 22 33 12&21 23&32 31&13
xx yy zz xy&yx yz&zy zx&xz

Similarly, the stress tensor V ij can be written as a pseudo-vector

V >V x V y V z W xy W yz W xz @T (2.17)
PRELIMINARIES 11

1.4 CONCLUSIONS
At the beginning of this chapter we have shown on an example in geomechanics that
substantial gains can be made with the BEM, in terms of mesh generation and solution
times. These gains are most pronounced for problems involving infinite or semi-infinite
domains. Other examples where the BEM seems to be superior to the FEM is for
problems where boundary stresses are important, e.g. in Mechanical Engineering.
Examples of this will be shown later.
The main purpose of this book is to encourage the use of the method. The simple
computer programs included contain all the necessary building blocks for building more
advanced and more specific computer programs for research or industrial applications.
In conclusion the reader should see this book as an advanced introduction to the
BEM, with some basic building blocks for computer programming.

1.5 REFERENCES

1. Ritz, W. (1909) ber eine Methode zur Lsung gewisser Variations-Probleme der
mathematischen Physik. Journal fr reine und angewandte Mathematik, 135:1-61.
2. Trefftz, E. (1926) Ein Gegenstck zum Ritzschen Verfahren. Proc. 2nd int. Congress
in Applied Mechanics, Zurich, pp 131.
3. Zienkiewicz O.C. and Taylor R.L. (2000) The Finite Element Method - Fifth Edition.
Butterworth-Heinemann, UK.
4. Timoshenko, S.P. and Goodier, J.N. (1970) Theory of Elasticity. McGraw-Hill,
London.