Chapter 1

The wave equation

In this Chapter we will consider the one-dimensional and multi-dimensional forms
of what is often referred to as the wave equation. Although this equation is not the
only one that describes wave-like behaviour, and although its solutions do not possess
the more complicated properties of the dispersive and nonlinear equations we shall
consider in later chapters, it is the prototype for much of what follows. This equation is
also important in itself, because it describes the behaviour of many important physical
systems [see, e.g., 7.1 of Whitham (1974) for some of them]: we will concentrate on its
application to vibrations and sound waves.
As we will see, the convenient feature of the wave equation is that it is linear. This
means that we can construct the solutions to complicated boundary- or initial-value
problems by superposing simpler solutions. In particular, by Fourier decomposition
we can write any periodic solution as the superposition of sinusoidal signals with dif-
ferent frequencies. It follows from this that to understand the behaviour of systems
described by linear wave equations it is sufcient to understand the behaviour of these
individual modes.
1.1 One dimension: waves on a stretched string
One of the simplest physical systems that supports waves is a stretched string. We
consider a uniform string, with constant density (mass per unit length) , under a
tension T. If an element of this string is displaced laterally by a small distance , the
tensional forces acting on this element will tend to pull it back into alignment. Because
the string has mass, it has inertia, so instead of returning the string to its undisturbed
position, the restoring forces can lead to oscillation about this position, rather like the
simple harmonic oscillator that you may remember from earlier Mechanics classes.
Section 2.1 of Billingham & King (2000) gives a derivation of the governing equation
for the displacement (x, t) of the string. The equation that is obtained is generally
5
known as the wave equation in one dimension. It is given by

t
2
= c
2

x
2
, (1.1.1)
where c is a positive constant. (In the case of the stretched string, c =
_
T/ and
represents its lateral displacement. However, other physical systems also obey (1.1.1)
or similar equations; to emphasise the mathematical connections, when we are dealing
with linear waves we will generally use to denote the dependent variable, and we
will use c to denote a constant wavespeed, regardless of the physical mechanisms that
underlie the equation.)
1.1.1 Standing waves
One of the simplest problems we can consider is the motion of a stretched string that
is xed at each end. If the length of the string is L, we must seek solutions of (1.1.1)
subject to the boundary conditions
(0, t) = 0 and (0, L) = 0. (1.1.2)
(We will discuss initial conditions later.)
We begin by seeking separable solutions of the form
(x, t) = F(x)G(t). (1.1.3)
Substituting this into (1.1.1) and separating the variables we obtain
1
c
2
G

(t)
G(t)
=
F

(x)
F(x)
= k
2
, (1.1.4)
where we have looked ahead slightly by writing the separation constant as k
2
.
Integrating the equation in x now yields
F(x) = Acos(kx) + B sin(kx). (1.1.5)
The boundary condition (0, t) = 0 now requires that A = 0, while the boundary
condition (L, t) = 0 requires that sin(kL) = 0, and thus that k = m/L for some
m Z.
Integrating the equation in t, similarly, yields
G(t) = C cos(t) + Dsin(t), (1.1.6)
where the angular frequency = ck. Putting these components together and setting
B = 1 without loss of generality, the general form of our separable solution is
(x, t; m) =
_
C cos
_
mct
L
_
+ Dsin
_
mct
L
__
sin
_
mx
L
_
for m Z. (1.1.7)
6
-1
-0.5
0
0.5
1
0 0.2 0.4 0.6 0.8 1
-1
-0.5
0
0.5
1
0 0.2 0.4 0.6 0.8 1
x

Figure 1.1: Snapshots of (x, t; m) for m = 1 (top) and m = 2 (bottom), in units such
that L = 1, c = 1, C = 1 and D = 0. Snapshots are taken at t = 0 (solid lines) and
mct/L = 0.25, 0.5, 0.75 and 1 (dashed lines).
We note several interesting features of this standing wave solution.
First, the separable solution is not unique. In fact, we have an innite number of pos-
sible solutions corresponding to the possible values of m (the eigenvalues of the prob-
lem); these solutions are known as modes (or eigensolutions). Once we have picked
a value of m, this determines the wavenumber k = m/L and the angular frequency
= mc/L which characterise the mode.
1
(Note that the wavenumber is related to
the more familiar wavelength by k = 2/, so the condition that kL = m is simply
equivalent to requiring that L = m/2, i.e. that an integer number of half-wavelengths
t into the length of the string L.) We will encounter the wavenumber and wavelength
again when we consider travelling waves, but we will nd that in that case there is no
equivalent restriction on the possible values of k.
Another typical feature of standing waves is that each mode has both nodes (points
where the amplitude of the oscillation is zero) and antinodes (points where the ampli-
tude of the oscillation is maximum). For example, the standing wave in gure 1.1(b)
has a node at x = 0.5 and antinodes at x = 0.25 and x = 0.75.
A nal crucial point is that, having obtained the modes of the form (1.1.7), we can
now construct the solution to the governing equation (1.1.1), subject to the boundary
1
For a guitar string, the lowest mode, m = 1, would correspond to the fundamental frequency, i.e.
to the note that that string makes when plucked; the higher modes, or harmonics, are part of what give
a guitar its characteristic sound.
7
conditions (1.1.2), given any initial condition on . This is because of two facts. First,
(1.1.1) is linear, which means that any linear combination of solutions is itself a solution
to the equation, and this means that we can construct more complicated solutions by
superposing modes of the form (1.1.7). Second, by Fourier decomposition we know
that we can write any initial condition satisfying (1.1.2) as a Fourier series, in other
words as a superposition of these modes at t = 0. We can express this by saying that
a given initial condition excites a set of modes at t = 0, each of which then evolves
over time as predicted by (1.1.7). By understanding the behaviour of these individual
modes, we are able to describe the general behaviour of the system.
Example 1.1. Consider a string of length L, xed at each end, which is plucked by
drawing the midpoint of the string a distance h, holding it stationary for a moment,
then releasing it. The initial condition is thus
(x, 0) =
_

_
2h
L
x in 0 x
1
2
L,
2h
L
(L x) in
1
2
L x L.
To obtain the subsequent evolution of the string, we start by writing the initial condi-
tion as a Fourier series,
(x, 0) =

m=0
a
m
sin
_
mx
L
_
for some coefcients a
m
. We can determine these coefcients in the usual way. For any
integer n > 0, we have
_
L
0
(x, 0) sin
_
nx
L
_
dx =
_
L
0

m=0
a
m
sin
_
mx
L
_
sin
_
nx
L
_
dx,
and thus
_
L/2
0
2h
L
xsin
_
nx
L
_
dx +
_
L
L/2
2h
L
(L x) sin
_
nx
L
_
dx =
1
2
La
n
.
Evaluating the integrals on the left-hand side, we obtain
2h
L
2L
2
n
2

2
sin
_
n
2
_
=
1
2
La
n
,
and thus
a
n
=
_
_
_
8h
n
2

2
(1)
(n1)/2
when n is odd,
0 when n is even.
The full solution for (x, t) can now be obtained by superposing these modes, to obtain
(x, t) =

n odd
8h
n
2

2
(1)
(n1)/2
sin
_
nx
L
_
cos
_
nct
L
_
.
The coefcients a
n
are sometimes called the partial amplitudes. Since a
n
1/n
2
, as n
increases the amplitudes of the higher harmonics decay rapidly, and the sound that we
hear will be dominated by the fundamental frequency and the rst few harmonics.
8
1.1.2 Travelling waves
We now consider the solutions that are possible when we do not have boundary condi-
tions to satisfy: we can think of these as describing waves on an innitely long string.
The most general solution of this kind is dAlemberts solution,
(x, t) = f(x ct) + g(x + ct), (1.1.8)
where f and g are arbitrary twice-differentiable functions.
Although weve considered the wave equation to describe a genuinely one-dimensional
system, the terminology that is commonly used comes from situations where a wave
propagates in three dimensions but still obeys equation (1.1.1). In this case, f(x ct) is
constant at all points in a plane perpendicular to the x-direction (called a wavefront)
that moves with speed c in the positive x-direction. Hence f(x ct) is a travelling
plane wave travelling in the positive x-direction with constant speed c, independent
of f, with neither distortion (change of shape) nor attenuation (loss of amplitude).
Similarly, g(x + ct) is a travelling plane wave travelling with constant speed c, inde-
pendent of g, in the negative x-direction.
Harmonic plane waves
As we did with the standing waves, we can write dAlemberts solution, for any choice
of f and g, as a superposition of sinusoidal travelling waves, known as harmonic plane
waves. The most convenient form for a harmonic wave in one dimension is
(x, t) = a cos(t kx + ), (1.1.9)
where a, , k and are constants. Here a is the amplitude; t kx + is the phase; is
the constant part of the phase and is sometimes called the phase angle; || is the angu-
lar frequency (in rad s
1
), so T = 2/|| is the period; and k(> 0) is the wavenumber
(the number of waves per 2 metres), so = 2/k is the wavelength. Note that the
harmonic wave (1.1.9) represents a sinusoidal disturbance that travels with velocity
/k without distortion or attenuation, and that it will be a solution of (1.1.1) as long as

2
= c
2
k
2
. (This is a simple example of a dispersion relation, which we will meet again
in chapter 2.)
The complex representation of uniform harmonic plane waves
Note that a cos(tkx+) is just the real part of ae
i(tkx+)
. This suggests the following
complex representation for one-dimensional uniform harmonic plane waves:
(x, t) = Ae
i(tkx)
, (1.1.10)
9
where A is the constant complex amplitude; A = ae
i
so that a = |A| is the real ampli-
tude and = arg(A). As long as we are considering linear governing equations with
real coefcients, the real and imaginary parts of will each independently satisfy the
equations. (This is not, of course, the case for the nonlinear equations we will meet in
Chapter 3.) In this representation it is to be understood that the actual wave is the real
part.
The following result is useful when using the complex representation: if = Ce
i
(so
that physically = Re
_
Ce
i
_
) then

2
:=
1
2
_
2
0
(Re())
2
d =
1
2
CC

, (1.1.11)
where
2
is understood to mean the average over one cycle of the square of the physical
quantity Re() and

denotes the complex conjugate.
To prove (1.1.11) let C = p + iq, where p and q are real, so that Re() = p cos q sin .
Then

2
=
1
2
_
2
0
(p cos q sin )
2
d
=
1
2
_
2
0
(p
2
cos
2
+ q
2
sin
2
2pq sin cos ) d
=
1
2
(p
2
+ q
2
)
=
1
2
CC

.
Example 1.2. Consider a harmonic wave propagating on an innite string that has an
abrupt change in density, so that c = c
1
for x < 0 and c = c
2
for x > 0. At x = 0, we will
impose the boundary conditions that is continuous (so the string does not break) and
that /x is continuous (which will ensure that we do not have a nite force acting
on an innitesimal element of the string).
It is not obvious at rst what conditions we are free to apply. We will assume that
in x < 0 there is an incident wave of amplitude A and wavenumber k propagating
rightwards, corresponding to a displacement

I
(x, t) = Ae
i(tkx)
,
where k > 0 and
2
= c
2
1
k
2
.
When the incident wave passes the density change at x = 0, its wavespeed will change.
Consider the transmitted wave in x > 0: assuming this is also travelling rightwards, it
must have the form

T
(x, t) = A
T
e
i(
T
tk
T
x)
for some constants A
T
,
T
and k
T
> 0, where
2
T
= c
2
2
k
2
T
.
10
Now we attempt to satisfy the boundary conditions at x = 0. We have

I
(0, t) = Ae
it
,
T
(0, t) = A
T
e
i
T
t
,

I
x
(0, t) = ikAe
it
,

T
x
(0, t) = ik
T
A
T
e
ipt
.
The condition
I
(0, t) =
R
(0, t) immediately tells us that A
T
= A and
T
= . The
condition that
I
/x =
T
/x at x = 0 then requires that k
T
= k. However, since

T
= , this is only possible if c
1
= c
2
. Something has gone wrong!
To give ourselves the freedom to satisfy both boundary conditions, we need to consider
a third component of the solution, a reected wave in the region x < 0, which travels
leftwards and has displacement

R
= A
R
e
i(
R
t+k
R
x)
,
where k
R
> 0 and
2
R
= c
2
1
k
2
R
.
Including the reected wave, our boundary conditions become

I
(0, t) +
R
(0, t) =
T
(0, t), i.e. Ae
it
+ A
R
e
i
R
t
= A
T
e
i
T
t
and

I
x
(0, t) +

R
x
(0, t) =

T
x
(0, t), i.e. ikAe
it
+ ik
R
A
R
e
i
R
t
= ik
T
A
T
e
i
T
t
.
The rst of these conditions can be satised only if
T
=
R
= , which also xes
the wavenumbers k
T
= /c
2
and k
R
= /c
1
. Substituting these into the boundary
conditions yields
A + C = A
T
,

c
1
A +

c
1
A
R
=

c
2
A
T
.
Solving for the unknown amplitudes A
T
and A
R
, we obtain
A
T
=
2c
2
c
1
+ c
2
A and A
R
=
c
2
c
1
c
1
+ c
2
A.
Note that when c
2
= c
1
the amplitude A
T
of the transmitted wave is equal to the
amplitude A of the incident wave, while the amplitude A
R
of the reected wave is
zero. When c
2
> c
1
, the reected wave has the same phase as the incident wave, but
when c
2
< c
1
its phase is exactly opposite. Finally, the larger c
2
becomes relative to c
1
,
the greater the magnitude of the transmitted wave relative to that of the incident wave.
1.2 Two dimensions: waves on a stretched membrane
We will now generalise the wave equation from one dimension to two (and later three)
dimensions. It is useful to have a specic physical system in mind: when we look at
11
three-dimensional waves, we will consider sound waves in a gas, but in two dimen-
sions we will start with a natural generalisation of the stretched string. We consider
a plane membrane stretched to a uniform tension T, such as a drumskin. Again, this
membrane may deform very slightly out of the plane, with displacement (x, y, t). The
governing equation can be derived in a very similar way to (1.1.1) [see, e.g., 27 of
Coulson (1952)], and may be written as

t
2
= c
2
_

x
2
+

2

y
2
_
. (1.2.1)
Many of the properties of this equation are similar to those of the three-dimensional
wave equation that we will meet in the next section, so we will not explore them in de-
tail here. However, it is interesting to consider the standing-wave problem: in the same
way that the one-dimensional standing wave problem told us how to calculate the nat-
ural frequencies (the fundamental frequency and its harmonics) of a guitar string, this
will tell us how to calculate the natural frequencies of a drumskin.
Lets rst consider a rectangular membrane of length L and width W, where L W,
and choose our co-ordinates so that the boundary conditions become
= 0 on x = 0, x = L, y = 0, and y = W. (1.2.2)
As in 1.1.1, we seek separable solutions to (1.2.1) that satisfy the boundary conditions
(1.2.2). It is easy (and a good exercise for you!) to show that the solutions have the
form
(x, y, t) = Ae
it
sin
_
mx
L
_
sin
_
ny
W
_
, (1.2.3)
where m and n are integers and where must satisfy the dispersion relation

2
= c
2

2
_
m
2
L
2
+
n
2
W
2
_
. (1.2.4)
The lowest frequency at which the membrane can vibrate is

1,1
= c
_
1
L
2
+
1
W
2
_
1/2
. (1.2.5)
In contrast to the guitar string problem, however, the next highest frequency will be

2,1
= c
_
4
L
2
+
1
W
2
_
1/2
, (1.2.6)
and this is not an integer multiple of the lowest frequency. In musical terms, the over-
tones made by a drum are not harmonics of the fundamental frequency, and this ex-
plains why drums make a much less pure sound than most other instruments.
12
Digression 1.1. We can carry this question a little further. Suppose we know the physical
properties of the membrane, and thus that we know c. If we can then measure the two low-
est frequencies that the drum produces, the expressions for
1,1
and
2,1
will give us enough
information to deduce the values of L and W. Thus, we can hear the shape and size of the
drum.
A much trickier question is whether we can do this in general: suppose that we know c, but
we dont know anything about the shape of the drum in particular, we no longer know that
its rectangular. If we listen very carefully, i.e. if we can determine the frequencies of all the
overtones, can we tell what the shape of the drum is? This question was posed by Mark Kac
[Can One Hear the Shape of a Drum? American Mathematical Monthly 73(4): 123, 1966],
and it wasnt answered until 1992, using some rather sophisticated mathematics to do so! The
Wikipedia article Hearing the shape of a drum gives a nice introduction to the problem.
1.3 Three dimensions: sound waves
A derivation of the governing equations for sound waves in an inviscid uid can be
found in section 3.1 of Billingham & King (2000). Sound waves are small uctuations
in the pressure and density of the uid, so the basic idea of the derivation is as follows.
We take a stationary uid at constant density
0
and corresponding pressure p
0
(note
that the density is a function of the pressure, (p)). We then introduce a small perturba-
tion to the pressure and density, with a corresponding small velocity, and we use the
smallness of these perturbations to linearise the equations. The governing equation
that we eventually obtain is

t
2
= c
2

2
. (1.3.1)
Here
2
is the Laplacian operator in three dimensions, which can be written in terms
of Cartesian co-ordinates as

2
=

2
x
2
+

2
y
2
+

2
z
2
. (1.3.2)
The dependent variable represents the velocity potential: it is related to the velocity
u and the pressure perturbation p by
u = and p =
0

t
, (1.3.3)
where
0
is the undisturbed density of the uid. The wave speed c is given by
c =
_
d
dp

p=p
0
_
1/2
, (1.3.4)
where p
0
is the undisturbed pressure of the uid. In the atmosphere at sea level, c
340 ms
1
, while in water, c 1500 ms
1
.
13
As before, we can seek both standing-wave and travelling-wave solutions to (1.3.1).
We will in fact leave the standing-wave solutions for the Exercises, and go straight to
travelling-wave solutions, which are analogous to those we saw in 1.1.2.
1.3.1 Harmonic plane waves
As we did for the wave equation in one dimension, we will introduce some notation
and terminology that we will use again in the following chapter.
Consider the single dependent variable (x, t), where in two dimensions x = (x
1
, x
2
)
and in three dimensions x = (x
1
, x
2
, x
3
). In two or three dimensions the plane-wave
generalisation of (1.1.9) is
(x, t) = a cos(t k x + ), (1.3.5)
where a, and are constants, and k is the constant wavenumber vector with k =
(k
1
, k
2
) in two dimensions and k = (k
1
, k
2
, k
3
) in three dimensions. The wavenumber
is k = |k|.
The complex representation of (1.3.5) is
(x, t) = Aexp (i(t k x)) , (1.3.6)
where A is a constant [cf. (1.1.10)].
The quantity
= t k x (1.3.7)
is the phase. The phase surfaces = constant are parallel planes. As = k, the
wavenumber vector k is perpendicular to these planes.
For waves governed by equation (1.3.1), it is easy to obtain the dispersion relation

2
= c
2
k
2
= c
2
_
k
2
1
+ k
2
2
+ k
2
3
_
. (1.3.8)
1.3.2 Reections and waveguides
Consider what happens when a harmonic plane wave meets a rigid wall. Well refer to
the original wave as the incident wave, and choose co-ordinates such that the wall is
at y = 0 and the wavevector is k = (k
1
, k
2
, 0), where k
1
> 0 and k
2
< 0, and where the
frequency satises
2
= c
2
(k
2
1
+ k
2
2
). The incident wave thus has velocity potential

I
(x, y, t) = Ae
i(tk
1
xk
2
y)
. (1.3.9)
Since there is a rigid wall at y = 0 with normal n = e
y
, we must impose the no-
penetration boundary condition u n = 0 on y = 0. This becomes

y=0
= 0. (1.3.10)
14
Clearly (1.3.9) does not satisfy this boundary condition. To satisfy it, we assume the
existence of a reected wave with velocity potential
R
, so that =
I
+
R
. We set

R
(x, y, t) = Be
i(
R
tk
R1
xk
R2
y)
, (1.3.11)
where the frequency
R
and wavevector k
R
= (k
R1
, k
R2
, 0) of the reected wave are to
be determined.
Substituting =
I
+
R
into (1.3.10) yields
k
2
Ae
i(tk
1
x)
k
R2
Be
i(
R
tk
R1
x)
= 0, i.e. k
2
Ae
i(
R
)t
= k
R2
Be
i(k
1
k
R1
)x
.
(1.3.12)
The LHS is independent of x and the RHS is independent of t, so we conclude that

R
= and k
R1
= k
1
.
Since
R
satises the wave equation (1.3.1), we also know that
2
R
= c
2
(k
2
R1
+ k
2
R2
), and
we conclude that k
2
R2
= k
2
2
. This leaves us with two possibilities. If k
R2
= k
2
then
(1.3.12) gives B = A, and
R
=
I
: we have rediscovered the zero solution to the
wave equation. There is a less trivial solution, however, when k
R2
= k
2
, so (1.3.12)
yields B = A. Physically, the reected wave has the same amplitude as the incident
wave; its phase on the wall y = 0 is the same as the incident wave, but the y-component
of its wavevector is reversed. In other words, it is reected exactly as one might navely
have expected.
Example 1.3. Consider a waveguide which consists of two parallel rigid walls, at y = 0
and at y = d. What extra restrictions does this place on travelling waves that propagate
in the space between the walls?
[This example will be discussed in class.]
1.3.3 Point sources
A nal type of solution to the three-dimensional wave equation deserves a brief com-
ment. Consider waves generated by a point source at the origin (an idealised loud-
speaker). To describe such waves, its convenient to work in spherical polar co-ordinates
(r, , ), so the wave equation (1.3.1) becomes

t
2
= c
2
_
1
r
2

r
_
r
2

r
_
+
1
r
2
sin()

_
sin()

_
+
1
r
2
sin
2
()

2
_
. (1.3.13)
Since the problem is isotropic (i.e. there is no preferred direction), we can reasonably
assume that the solutions should be spherically symmetrical, so is independent of
and , and the governing equation reduces to

t
2
= c
2
1
r
2

r
_
r
2

r
_
. (1.3.14)
15
At this point a trick presents itself. We dene a new dependent variable = r, and
we can then rewrite the RHS of this equation rather nicely as
1
r
2

r
_
r
2

r
_
=
1
r
2

r
_
r
2

r
_

r
__
=
1
r
2

r
_
+ r

r
_
=
1
r

r
2
. (1.3.15)
We can therefore write (1.3.14) as

t
2
= c
2

r
2
. (1.3.16)
In other words, the quantity = r satises the one-dimensional wave equation. In
particular, there are solutions of the form
(r, t) = f(r ct) , i.e. (r, t) =
1
r
f(r ct) (1.3.17)
for any twice-differentiable function f. Such solutions correspond to an outward-
propagating wave which maintains its shape (described by f) but decays in amplitude
as 1/r with distance from the source.
Digression 1.2. You may recall that f(r ct) was only one part of dAlemberts general so-
lution (1.1.8) to the wave equation. In principle, then, we could nd mathematically valid
spherically symmetrical solutions of the form (r, t) = (1/r)g(r + ct). However, these so-
lutions would not generally be regarded as physically meaningful, because they correspond to
an identical signal being transmitted towards the origin from innitely far away, from every
direction simultaneously.
A more general way to look at this is to say that physically valid solutions to the wave equa-
tion should satisfy a radiation condition. Such a condition states that energy (or in some
formulations information) should not propagate into the system from innity, though energy
(or information) can be radiated outward towards innity. In turn, this condition can often
be justied, and in some cases formulated more precisely, by requiring that physical solutions
should originate from an initial-value problem where the disturbance is originally conned to
a nite region. [See 6.7 of Stoker (1957) for a discussion of this point.] In this case, we can
think of the radiation condition as a way of imposing causality.
We will think a little more about both energy and information in the following chapters...
16