RESEARCH NOTES

Boundary element method and wave

equation
Djuro M. Misijenovic
Mathematical Institute, Beograd, Yugoslavia
(Received Jarmary 1982)

In this paper the boundary integral expression for a one-dimensional wave equation
with homogeneous boundary conditions is developed. This is done using the time
dependent fundamental solution of the corresponding hyperbolic partial differential
equation. The integral expression developed is a generalized function with the same
form as the well-known D’Alembert formula. The derivatives of the solution and
some useful invariants on the characteristics of the partial differential equation are
also calculated.
The boundary element method is applied to find the numerical solution. The
results show excellent agreement with analytical solutions.
A multi-step procedure for large time steps which can be used in the boundary
element method is also described.
In addition, the way in which boundary conditions are introduced during the time
dependent process is explained in detail. In the Appendix the main properties of
Dirac’s delta function and the Heaviside unit step function are described.

Key words: boundary integral element method, wave equation

Introduction electromagnetic, etc. The equation presents two types of

boundary conditions:
The Boundary Integral Element method has become an
increasingly popular method for solving problems described ll=ii on K2;lxR (24
with partial differential equations.’ The first analytical au
results for boundary integral methods were developed at an’4 on KJxR (2b)
the beginning of this century. The method was re-examined
10 years ago using the new technology which can be effi- We assume that functions ii and Q are equal to zero, i.e. the
ciently applied for approximate solutions.2-4 Currently, boundary conditions are homogeneous.
many researchers are interested in the boundary element Also, equation (1) requires the following initial condi-
solution of time dependent problems. Boundary element tions:
results have been represented for parabolic partial differen-
tial equationF7 but for hyperbolic partial differential u = @I onax{ @a)
equations the theory and application are not as well all
developed.*-” - zr2 onfix{ (3b)
at-
Here, without loss of generality, we assume the time interval
Formulation of the problem as [0, T]. Ln a more general case we may have to perform
We want to sblve the following one-dimensional hyperbolic shifting (translation in time domain). Also, we suppose that
partial differential equation: $2 is finite set, n = [0, L].

a% a5
---2 - = 0 xE52,tER (1) Developing BIEM expression
at* ax*
We interpret equation (1) in terms of potential theory, i.e.
This is the well-known wave equation which governs many u is a potential and au/an a flux along the outward-directed
problems in continuum mechanics, such as acoustical, normal at the boundary point under consideration.

0307-904X/82/030205-04/$03.00
0 1982 Butterworth & Co. (Publishers) Ltd. Appl. Math. Modelling, 1982, Vol 6, June 205
Research Note

In potential theory, the adjoint equation to equation It is easy to see that the first integral term in equation
(1) is: (lo), after applying (4) and the selective property of the
delta function, gives:
a3.d
- - $ a’” = -6(x -J,) &(r - 7) (4) L
at2 a2 all all* =+E
11(x, f) = l.P--U-- dy (11)
which expresses the effect of unit source at the point 01, r) S( at at )I o
0
in the absence of other source points in the domain of
interest. The asterisk (*) denotes the unbodnded character Now, introducing the causality property of the funda-
of this solution. c is the speed of propagation of the disturb- mental solution, i.e.:
ance in the medium under consideration. 6(a) is described
in the Appendix. ~(*0,,7;x,t)=O wheneverclr-tl<ly--x’l (12)
It is well known that the so-called fundamental solution
and the explicit formulation of the function cl*, as given by
for equation (4) is given by:
(9) we obtain:
u*=--fqct-r] (5) L
2c I a
u(x, t)= -- --H[cr-lly--lluOt,O)dy
H(o) is described in the Appendix, and r is the distance s 2~ at
0
between source point and point under consideration.
We transform equation (1) to a corresponding integral L
expression using the weighted residual method, i.e. multi- a4.h 0)
plying equation (1) by a function w, which gives: + ardy (13)

Using expression (A5) and the ‘cut off’ property of the

w dy dr = 0 (6) Heaviside function we have the following formulation:
0 0

where w is an arbitrary function to be defined. 11(x, t) = f [ll(x - ct, 0) + u(x + ct, O)]
After partial differentiation of expression (6) and
introducing homogeneous boundary conditions, we obtain:
(14)

This expression is understood in the sense of a Cauchy

principal value and it allows for the solution of a much
wider class of problem than the ordinary expression known
as the D’A1amber-t formula.
It is clear from the above that expression (14) is also
valid on the upper boundary t = T of the domain of
As is usual in the weighting residual method, the function interest. There, we have:
w distributes errors over the whole domain.
The best choice for the weighting function w is to take: u(x, T) = +(x -cT,O) + u(x + cT,O)]
w=u* (8)
Function w is a function of four variables: the first two
variables are the space-time coordinates of the source (15)
point and the second pair of variables correspond to the
coordinates of the point under consideration. This func-
tion is given by: Note that all integrals are now given on the boundary (some
of them are given as functional values and the rest of them
disappear because of homogeneous boundary conditions).
w=w(y,T;X,t)=-~H[c,T-t/-ly-xll (9)

With this choice of weighting function expression (7) gives: Calculation of derivatives of the solution and some
useful invarian ts
T+E L
a2u* 2az”* Expression (14) (or equivalently (15)) can be used for
--C u dy dr calculating the partial derivatives of the solution with
at2 ax2
respect to time and space. They are:
L
a+ c - au+ - ct, 0) + a+ + ct, 0)
+
I(
0
au
u*--u
at
au*
- T+E
at )I o
dy = 0 (10)
-=-
at
t)

2 [ ax ax 1
I au(x - ct, 0) + al+ + ct, 0)
Next, we integrate in time in the interval [0, T + E], E > 0.
+-
2 at at 1 (16)

206 Appl. Math. Modelling, 1982, Vol 6, June

 Research Note
and numerical results oscillate around the exact solution and
the large number of points are needed to reduce this ‘noise’.
I au(x - ct, 0) + ~U(X + ct, 0)
1
&1(x, t)
-=- The boundary element solution on the other hand did not
ax 2 ax ax have this problem and results were coincident with the
exact ones.
+ L _ au(x - ct, 0) + au(x + ct, 0)
2c [ at at 1
(17)

Note the simplicity of equations (16) and (17) which Multis tep procedure
depends only on functional values of the corresponding A problem of practical importance is to calculate a solution
partial derivatives. with time steps as large as possible. In this case we can use
From the above expressions we can conclude that the the following multistep procedure.
following two quantities are constant in the direction of the Let CT = s, where s = length of one element. We want to
characteristics: calculate the solution at the time t = 2T. Shifting time by
one time step in expression (15) gives:
au(x, t) aLc(x, t>
c-++
ax at
u(x, 2T) = f [u(x -s, T) + u(x + s, T)]
aL[(x + d, 0) + au(x + ct, 0)
=c (18)
ax at
T x;s au(y, T)
+-
2s J ~
au(x, t)
c--- aflcx, t) dv (21)
at
ax at x-s
aucx - ct, 0) au(x - ct, 0) After introducing (1.5) (16) and (17) we find:
=c (19)
ax - at
u(x, 2T) = ; [u(x - 2s, 0) + u(x + 2s, 0)]
Equation (18) and (19) gives the characteristic directions of
propagation of disturbance in the space-time domain.
x+2&!
auk 0)
+f - dy cm
Boundary element method formulation 4s s at
x-2s
We use expressions (1 S), (16) and (17) for numerical calcu-
lations. !2 is divided in N equal parts. We describe each part This formula can, by induction, be extended to:
with two nodes (end points) and three values-degrees of
freedom (u, au/ax, au/at) at each node. Hence, we have u(x, IIT) = ; [u(x -

.Y+fl.T
IIS, 0) + u(x + IZS, O)]
cubic variation of the function cl in the space domain. All
nodes (except the first and last) were double nodes. This
approach allows the discontinuity of the space derivative of
the solution to be taken into consideration. In the case for
a shape that is continuous in the space domain, any dis-
+-Tadi
0)
s dy,
~ 2ns at
nEN (23)

continuity of the derivative will indicate a numerical error.

Numerical examples Introducing boundary conditions

We have solved two casesof a vibrating string with fixed If we want to have the same formula for calculating the
ends. In both cases, the string had unit length and the speed position of all the points in the string including the two
c was chosen to be equal to 10. boundary points, we need to define some values outside the
In the first example, the displacement is 0.1 in the string domain. In wishing to do so, we need to assume that
middle of the string and varies linearly towards the fixed the string extends outside R and that formula (15) can be
ends. applied in the following way.
In the second example the 0.1 displacement was made to At the initial moment the integrand of (15) is equal to
act at a point located one-third of the way along the string zero and the displacement on the left-hand side (fwed end)
length. These were the two initial positions of the string. is equal to zero. As we have no restrictions on time in
Initial velocities were assumed in both cases to be zero. this formula, it follows for this end of the string that:
Numerical results with only six elements give the same u(-z, t) = - u(z, t) z E [0, s], t E [0, T] (24)
values as the exact solution.
The time step was given by the Courant number, K: In the same way, if the string passesthrough the equili-
brium position u(x,O) = 0, x E [0, L] and has some non-
c. At zero velocity we can conclude for fixed end x = 0 that:
KE ---cl
n ._ (20)
a+-z, t) auk t)
This time step corresponds to the time for which the _ =-- zE[O,s],tE[O,T] (25)
at at
disturbance advances across one whole element.
If the problem is solved using finite differences the solu- Similarconclusions are valid for the other fixed end at
tion progress without dumping or amplification but x=L.

Appl. Math. Modelling, 1982, Vol 6, June 207

Research Note

Concluding remarks 9 Brebbia, C. A. (ed.). ‘Boundary element methods’, Springer-

VerIag, Heidelberg, 1981
The problem which is under current investigation is the 10 Brebbia, C. A. (Private communication)
solution of shock waves. This is a more difficult problem
as it corresponds to nonhomogeneous boundary conditions
but encouraging results have already been obtained.
Appendix
Acknowledgements , Here, we briefly describe the properties of the Dirac delta
function and Heaviside unit step function which are used in
The author is grateful to Dr C. A. Brebbia, University of this paper. These functions are of great importance in
Southampton, for his helpful comments and support in all developing BIEM formulae.
stages of developing these results. 6(or) is the well-known Dirac delta function with two
He also thanks the Computational Mechanics Centre, main properties:
Southampton, for use of their computing facilities on
VAX 1 l/750, and the Mathematical Institute, Beograd, 6(a) = 0 a#0 (Al)
Yugoslavia, for every support. and
+-
References
Brebbia, C. A. ‘The boundary element method for engineers’,
Pentech Press, London, 1978
Brebbia, C. A. and Wrobel, L. C. ‘Recent advances in numerical
s 6((u) do = 1 (A21

methods in fluids’, VoL 1 (ed. Taylor, C. and Morgan, K.), The selective property of Dirac’s function gives:
Pineridge Press, UK, 1980, pp 1-25. +-
Wrobel, L. C. and Brebbia, C. A. ‘Boundary elements in water
resources’, NATO Instituto at Viierio. Portugal. Aunust 1981 (A31
Brebbia, d. A. and Wrobel, L. ‘Apphcations of boundary ele.
ments in water resources’, 3rd Int. ConJ Fin&e Ekrnerm in
Water Resour., 19-23 May, 1980, University of Mississippi,
Oxford, USA, pp 2.3-2.14 H(a) is the Heaviside unit step function defined by:
Shaw, R. P. ‘An integral approach to diffusion’, Inf. J. Hear
Mass Transfer 1914, 17, 693 (Y>O
Liggett, J. A. and Liu, P. I. F. ‘Unsteady flow in confined H(a)= :, C.44)
aquifers: a comparison of two boundary integral methods’, t a<0
Water Res. Res. 1979, 15 (4), 861 and with the useful property that:
Brebbia, C. A. and Walker, S. ‘Boundary element techniques
in engineering’, Newnes-Butterworth, 1380
Brebbia, C. A. (ed.). Progress in boundary elements’, VOL 1, $H(o!)= 6(a) (A9
Pentech Press, London, 1981

208 Appl. Math. Modelling, 1982, Vol 6, June