MATHEMATICS 0054 (Analytical Dynamics)

YEAR 2019–2020, TERM 2

HANDOUT #9: INTRODUCTION TO WAVES

1 What is a wave?
In everyday life we are familiar with lots of examples of waves: ripples on the surface of a
pond; a traveling wave along a rope; the oscillations of a violin string or an organ pipe; sound
waves in the air; electromagnetic waves (i.e., light, radio, etc.); and many other examples.
But it surprisingly difficult to give a general definition of what one means by a wave!
The main idea to keep in mind is that a wave is a pattern, not a thing; usually it is a
pattern that moves or changes in some way.1 Of course, this pattern is usually formed (or
“carried”) by the motion of some “medium”: for instance, the ripples on the surface of a
pond are formed by the motion of water molecules; a sound wave is formed by the motion of
molecules in the air; and so forth. But the motion of the wave can be very different — both
in direction and in magnitude — from the motion of the medium that “carries” the wave:

• When we drop a stone into a placid pond, a wave propagates concentrically outwards
along the surface of the pond (in principle to infinity, and in practice for many metres).
But the water molecules move up and down (and only a few millimetres).

• When a person speaks, a sound wave propagates concentrically outwards in three-

dimensional space (sometimes to hundreds of metres). But the molecules in the air
oscillate back and forth, and only very small distances (comparable to the typical
separation between molecules in the air).

Here are two even more bizarre examples:

• Imagine a queue to buy tickets at the cinema. When the person at the front of the
queue finishes buying his ticket, he departs from the queue, leaving an empty space
at the front. Then the second person moves into this empty space, leaving in turn an
empty space in position #2; then the third person moves into position #2, leaving an
empty space in position #3; and so forth. The pattern here is the empty space: it
moves backwards (quite possibly for a long distance). The medium that carries this
pattern is the group of people: they each move forwards (but only one step).

• In a Mexican wave in a stadium, successive rows of spectators stand, raise their arms,
and then sit again. This give rises to a pattern (standing spectators) that moves
horizontally through the stadium (quite possibly for a long distance). But the medium
that carries this pattern, namely the spectators, simply move up and down (one metre
1
For this point I am indebted to Ralph Baierlein, Newton to Einstein: The Trail of Light (Cambridge
University Press, Cambridge–New York, 1992), p. 60.

1
 or so) without leaving their seats. [This is of course completely analogous to a ripple
on a pond.]

Waves can be classified as longitudinal or transverse, according as the motion of

the medium is parallel or perpendicular to the direction of propagation of the wave. For
instance, sound waves are longitudinal: the molecules in the air oscillate back and forth
along the direction of propagation of the wave. By contrast, waves on the surface of water,
or traveling waves along a rope, are transverse: the molecules in the medium oscillate up and
down while the wave propagates horizontally. Electromagnetic waves are also transverse: the
electric and magnetic field vectors are perpendicular to the direction of propagation (and
also perpendicular to each other). Waves on a slinky can be either longitudinal or transverse
(that is one of the reasons that a slinky is so much fun to play with!).
Waves in free space are typically propagating waves: a pattern propagates in some
way (perhaps in one direction, perhaps concentrically outwards); the pattern may or may
not change in shape as it propagates.2 On the other hand, when waves are constrained by
boundary conditions — for instance, a violin string tied down at both ends, or an organ
pipe closed at one or both ends — it is possible to set up standing waves: these are fixed
shapes whose amplitude oscillates in time, so that the wave f (x, t) is of the simple product
form g(x)h(t) [where h(t) is usually a simple oscillation A cos ωt, while g(x) is the normal
mode].

2 Waves on an elastic string

Let us now derive a differential equation for the transverse motion of a string, in the lin-
earized (i.e. small-amplitude) approximation. The equation will describe the vertical height
y(x, t) of the string at point x along the string and time t. Since y depends on both x and t,
we will have a partial differential equation.
We work in the absence of gravity; therefore, “horizontal” is simply a synonym for “lon-
gitudinal” (i.e. along the direction of the string), and “vertical” is simply a synonym for
“transverse” (i.e. perpendicular to the direction of the string).

2.1 Direct derivation

We assume that the string has a constant linear density (= mass per unit length) ρ and
is held at a tension T . Because of our small-amplitude approximation, we may assume that
the tension T does not change as the string oscillates. Let us now consider a small segment
of the string, of width δx, centered at x, as shown in the figure below:
2
We will see that the one-dimensional wave equation has solutions that propagate to the left or right
without changing their shape. On the other hand, for the wave equation in two or more spatial dimensions,
solutions do usually change their shape as they propagate.

2
 The mass of this segment of string is ρ δx. Consider now the forces acting on this segment
of string. The string to the right of this segment pulls on it with a horizontal force T cos ψ2
and a vertical force T sin ψ2 , where ψ2 is the angle that the tangent to the string makes with
the horizontal axis at the rightmost end of our segment. Likewise, the string to the left of
this segment pulls on it with a horizontal force −T cos ψ1 and a vertical force −T sin ψ1 ,
where ψ1 is the angle that the tangent to the string makes with the horizontal axis at the
leftmost end of our segment. The net vertical force on this segment is therefore
Fvert = T sin ψ2 − T sin ψ1 . (1)
Our small-amplitude approximation means that we assume the angles ψ1 , ψ2 are small (they
are exaggerated in the figure for clarity); therefore sin ψ1 ≈ ψ1 and sin ψ2 ≈ ψ2 , and hence
in this approximation we have
Fvert = T (ψ2 − ψ1 ) . (2)
On the other hand,
∂y
(x + δx/2, t) = tan ψ2 ≈ ψ2 (3a)
∂x
∂y
(x − δx/2, t) = tan ψ1 ≈ ψ1 (3b)
∂x
Now3
∂y ∂y ∂2y
(x + δx/2, t) − (x − δx/2, t) = (x, t) δx + O((δx)2 ) , (4)
∂x ∂x ∂x2
so the net vertical force on this segment of string is
∂2y
Fvert = T (x, t) δx + O((δx)2) . (5)
∂x2
3
Here we have just used the Taylor expansion f (x + ǫ) = f (x) + ǫf ′ (x) + O(ǫ2 ) for the function f = ∂y/∂x
and ǫ = ±δx/2; the variable t just goes for the ride.

3
By Newton’s second law, the net vertical force on this segment should equal its mass ρ δx
multiplied by its vertical acceleration of its center of mass. But the vertical acceleration of
every point in this small segment is essentially the same as the vertical acceleration of its
∂2y
center point, which is 2 (x, t). (More precisely, different points within this segment will
∂t
have vertical accelerations that differ from this by an amount of order δx.) We therefore
have  2
∂2y

∂ y
(ρ δx) (x, t) + O(δx) = T (x, t) δx + O((δx)2 ) . (6)
∂t2 ∂x2
Dividing through by δx and letting δx → 0, we obtain the partial differential equation

∂2y ∂2y
ρ = T . (7)
∂t2 ∂x2
p
It is convenient to define c = T /ρ (we will see later that c is the speed of propagation of
waves along this string); we therefore have

∂2y 2
2 ∂ y
= c . (8)
∂t2 ∂x2
Remark. The net horizontal force on this segment is

Fhoriz = T cos ψ2 − T cos ψ1 , (9)

which is zero in the small-amplitude approximation (i.e., expanding everything through

linear order in the angles ψ) because to this order we have cos ψ1 ≈ 1 and cos ψ2 ≈ 1.
So there is no horizontal motion.

Let us henceforth rename the dependent variable y as f , in order to stress that the “thing
that oscillates” need not be the height of a string; it could be the pressure in a sound wave,
the voltage in an electrical transmission line, or many other things. The partial differential
equation
∂2f 2
2 ∂ f
= c (10)
∂t2 ∂x2
for the function f (x, t) is called the wave equation in one spatial dimension — or the
one-dimensional wave equation for short. Sometimes we prefer to divide this equation
by c2 and write instead
∂2f 1 ∂2f
= . (11)
∂x2 c2 ∂t2
And sometimes we write this equation in the form
 2
1 ∂2

∂
− 2 2 f = 0 (12)
∂x2 c ∂t

∂2 1 ∂2
in order to stress the role played by the one-dimensional wave operator − 2 2.
∂x2 c ∂t

4
 More generally, the wave equation in n spatial dimensions is

1 ∂2f
∆f = (13)
c2 ∂t2
where
∂2 ∂2 ∂2
∆ = + + · · · + (14)
∂x21 ∂x22 ∂x2n
is the Laplacian in n spatial dimensions (also denoted ∇2 ). It turns out that the wave
equation in n ≥ 2 spatial dimensions is much more interesting and subtle than the case
n = 1 that we are studying here! But understanding the solution of the one-dimensional
problem is a good warm-up for studying the more difficult multidimensional problem.

2.2 Alternate derivation as limit of discrete system

The equation (7) for the motion of the constant-density string can also be derived in a
slightly different way, by starting from the equations of motion of the massless string loaded
with discrete masses (derived in Section 5 of Handout #8) and taking the “continuum limit”.
Recall the situation we treated there: A massless string of length L = (n + 1)d is tied down
at the two ends x = 0 and x = L and held at tension T ; on this string we attach n particles,
each of mass m, at the locations x = d, 2d, 3d, . . . , nd. So the jth particle (j = 1, 2, . . . , n)
has longitudinal location x = jd; we denote its transverse displacement by yj .
We now consider the continuous string of linear density ρ to be a limit of the discrete
loaded string in which we take d → 0 and m → 0 with the ratio d/m = ρ held fixed.4 By
equation (44) of Handout #8 together with Newton’s second law, the equation of motion for
the discrete loaded string (in the linearized approximation) is

T
mÿj = (yj−1 − 2yj + yj+1 ) . (15)
d
We now assume that the yj are the values of a smooth function y(x) evaluated at the points
x = jd. Applying Taylor’s formula

f (x + ǫ) = f (x) + ǫf ′ (x) + 21 ǫ2 f ′′ (x) + O(ǫ3 ) (16)

with f = y, x = jd and ǫ = d, we obtain

∂2y
yj−1 − 2yj + yj+1 = d2 (x, t) + O(d3) (17)
∂x2
— that is, the “second difference” yj−1 − 2yj + yj+1 is a discrete approximation to the second
derivative ∂ 2 y/∂x2 (up to a factor d2 ). On the other hand, ÿj is simply ∂ 2 y/∂t2 . Since
m = ρd, we have
∂2y ∂2y
ρd 2 = T d 2 + O(d2) . (18)
∂t ∂x
Dividing by d and taking the limit d → 0, we obtain the one-dimensional wave equation (7).
4
If the length L of the string is fixed, this means that n → ∞.

5
 Note that in this derivation we nowhere used the length L of the string or the boundary
conditions at the ends. Indeed, in this section we are simply deriving the differential
equation (7) obtained by applying Newton’s laws of motion to the interior elements of
the string. The same differential equation (7) holds irrespective of whether the string
has boundaries or not, and irrespective of what goes on at those boundaries if they
do exist. Rather, if the string does have boundaries, then the differential equation has
to be supplemented by boundary conditions — we will discuss this aspect of the
problem in Section 4 below.

The connection seen here between the discrete loaded string and the continuous massive
string is prototypical of many similar situations arising in both pure and applied mathemat-
ics. A continuous problem can frequently be understood as a “continuum limit” of a discrete
problem when some sort of “lattice spacing” tends to zero: for instance, the derivative f ′ (x)
is the limit of a difference quotient [f (x + ǫ) − f (x)]/ǫ when ǫ → 0; the Riemann integral
is the limit of Riemann sums; and so forth. This connection between a continuous problem
and a sequence of discrete problems can be exploited in various directions:
• Sometimes we use the connection for theoretical purposes. For instance, we might
prove upper or lower bounds on a sum by relating it to an integral; this can be useful
if the sum is difficult to perform but the integral is easy.5 Or we might find that the
best way to make sense of an ill-defined continuum expression is to write it as a limit
of well-defined discrete expressions (this situation arises, for instance, in quantum field
theory).
• Sometimes it is the continuous problem that is of interest to us, but we study the
discrete problem (for n finite but large) as an approximation. For instance, if we want
to solve a differential equation (ordinary or partial) numerically on the computer, we
must discretize it, replacing derivatives by finite differences. (It is then prudent to
perform the numerical solution for at least two different values of the “lattice spacing”
ǫ, in order to test numerically whether the solutions seem to be almost converged.)
• Sometimes it is the discrete problem that is of interest to us, but we start by studying
a closely related continuous problem because it is easier to analyze. For instance,
differential equations are often simpler to analyze than difference equations, especially
if they are nonlinear.6
You will see many examples of these techniques in your future study of mathematics.
5
In analysis (MATH0003) you may have seen the example of the harmonic sum
1 1 1
Hn = 1 + + + ...+ ,
2 3 n
which is not given by any simple explicit
Z n formula but which can be bounded above and below in terms of
1
the corresponding definite integral dt, which is easily computable (of course it equals log n).
1 t
6
For instance, the logistic differential equation
dy
= y(1 − y)
dt

6
3 Solutions of the one-dimensional wave equation in
infinite space
In this section we shall find the general solution of the one-dimensional wave equation (11)
when the spatial variable x ranges over the entire real line — that is, for an infinitely long
string, without boundaries.

3.1 Guessing the general solution

We begin by observing that the one-dimensional wave equation admits traveling-wave
solutions of arbitrary shape, which move either to the right or to the left at speed c. To see
this, let F be any twice-continuously-differentiable function of a single real variable, and let
us consider the function
f (x, t) = F (x − ct) . (19)
Physically, this represents a pulse that has shape F (x) at time 0 [since f (x, 0) = F (x)]
and that moves to the right at speed c without changing its shape. (You should make sure
you understand why this pulse moves to the right as time goes on, i.e. moves towards larger
positive x.)7 Now compute the first and second partial derivatives of this f (x, t) with respect
to x:
∂f
= F ′ (x − ct) (20a)
∂x
∂2f
= F ′′ (x − ct) (20b)
∂x2
(why?). By similar reasoning we may differentiate with respect to t:
∂f
= −cF ′ (x − ct) (21a)
∂t
∂2f
= c2 F ′′ (x − ct) (21b)
∂t2
has applications in population biology and many other fields; and it is very easy to solve:
1
y(t) =
1 + e−(t−t0 )
where t0 is the time at which y = 12 . (You should derive this solution: note that the equation is separable.)
On the other hand, the discrete-time logistic map
yn+1 = ryn (1 − yn ) ,

where r ∈ [0, 4] is a parameter — which also has applications in population biology — has incredibly
complicated behavior for many values of the parameter r. This example shows that in some cases the
simplified continuous model is often a bad approximation to the discrete model; it all depends on what
aspects of the behavior one is interested in.
7
To keep at the same point along the pulse, x − ct must be kept constant. Therefore, as t increases, the
value of x must also increase in order to stay at the same value of x − ct. Indeed, if t increases by ∆t, then x
must increase by c∆t in order to stay at the same value of x − ct; so the pulse moves to the right at speed c.

7
(why?). It follows that f (x, t) = F (x − ct) is a solution of the one-dimensional wave equa-
tion (11), and this no matter what the function F is, provided only that it is smooth (i.e.
twice continuously differentiable).
By an analogous argument, the function

f (x, t) = G(x + ct) (22)

is also a solution (whenever G is twice continuously differentiable); it corresponds physically

to a pulse that moves to the left at speed c.
And by linearity, the sum of two solutions is also a solution: therefore,

f (x, t) = F (x − ct) + G(x + ct) (23)

is a solution of the one-dimensional wave equation (11). Physically, it corresponds to a pair

of pulses — one of shape F moving to the right, the other of shape G moving to the left —
that propagate without interacting with each other (i.e., the solution is just the arithmetic
sum of the two).8

Remark. Suppose that we had tried the more general Ansatz

f (x, t) = F (x − vt) (24)

where the velocity v is unspecified. Then we would have

∂2f
= F ′′ (x − vt) (25)
∂x2
and
∂2f
= v 2 F ′′ (x − vt) , (26)
∂t2
so that this f (x, t) solves one-dimensional wave equation (11) if and only if v 2 = c2 ,
i.e. v = ±c. (And this no matter what the function F is, provided only that it is twice
differentiable.) So this more general Ansatz would lead us back to the two solutions
v = c and v = −c that we found by clever guessing.9

8
The fact that the two pulses propagate without interacting with each other is, of course, a consequence
of the fact that we are dealing here with a linear differential equation, so that any linear combination of
two solutions is also a solution. The behavior of nonlinear partial differential equations is in general very
different!
9
There is actually another solution that we have missed here: Can you see what it is? The equation

v 2 F ′′ (x − vt) = c2 F ′′ (x − vt)

is solved by an arbitrary function F when v 2 = c2 ; but it is also solved for an arbitrary value of v whenever
F ′′ (x − vt) is identically zero, i.e. when F (x) = a + bx. But this is a rather bizarre solution from a physical
point of view, since the string’s displacement gets huge as x → ±∞.

8
3.2 Proof that it is the general solution
In fact, it turns out that (23) is the general solution to the one-dimensional wave equation,
i.e. every solution is of this form. To show this, let us make the change of variables from
(x, t) to new variables (ξ, η) defined by

ξ = x − ct (27a)
η = x + ct (27b)

And let us re-express the operators of partial differentiation with respect to x and/or t as
operators of partial differentiation with respect to ξ and/or η.
Warning! When computing partial derivatives arising from changes of variable in
functions of two or more variables, one must be very careful! The reason is this: when
∂f
we have a function f (x, y) and we write , we mean the rate of change of f as x varies
∂x
with y held fixed, but our notation fails to show this latter aspect; to be more precise
∂f 
we ought to write in order to make clear not only which variable is being varied
∂x y
but also which variable is being held fixed. Ordinarily this does not matter, because
we know implicitly which variable is being held fixed; but when we make changes of
variable this can become ambiguous. For instance, suppose that we change from y to
the new variable y ′ = y + x2 . We can write f either as a function of x and y or as a
∂f  ∂f 
function of x and y ′ . But 6= ! This fact causes endless complication (and
∂x y ∂x y′
confusion) in some applied subjects, such as thermodynamics.
In the case at hand, there will be no ambiguity: whenever we write partial derivatives
with respect to x and/or t, we assume that everything
 is being written as a function
∂ ∂ 
of x and t; so, for instance, means . Likewise, whenever we write partial
∂x ∂x t
derivatives with respect to ξ and/or η, we assume
 that everything is being written as
∂ ∂ 
a function of ξ and η, so that means . We never mix the two sets of variables.
∂ξ ∂ξ  η

If we increase x by ǫ with t being held fixed, this means that both ξ and η increase by ǫ;
therefore
∂ ∂ ∂
= + . (28)
∂x ∂ξ ∂η
Likewise, if we increase t by ǫ with x being held fixed, this causes η to increase by cǫ and ξ
to decrease by cǫ; hence
∂ ∂ ∂
= c − c . (29)
∂t ∂η ∂ξ
∂2 1 ∂2
The wave operator − can therefore be re-expressed in terms of ξ and η as
∂x2 c2 ∂t2
2 2
∂2 1 ∂2
 
∂ ∂ 1 ∂ ∂
− 2 2 = + − 2 c − c (30a)
∂x2 c ∂t ∂ξ ∂η c ∂η ∂ξ

9
 ∂2 ∂2 ∂2 ∂2 ∂2 ∂2
   
= + 2 + − − 2 + (30b)
∂ξ 2 ∂ξ∂η ∂η 2 ∂η 2 ∂ξ∂η ∂ξ 2

∂2
= 4 . (30c)
∂ξ∂η
So the one-dimensional wave equation can be rewritten in the variables ξ and η as

∂2f
= 0. (31)
∂ξ∂η
Now, what is the general solution to the partial differential equation (31)? That is, which
functions f (ξ, η) have the property that their mixed partial derivative ∂ 2 f /∂ξ∂η vanishes
identically? It is easy to see that any function of the form f (ξ, η) = F (ξ) + G(η) has this
property; and with a bit more thought one can see that only functions of this form have this
property.
Here is a proof of the latter statement. Let us rewrite the equation ∂ 2 f /∂ξ∂η = 0 as
 
∂ ∂f
= 0. (32)
∂ξ ∂η

This says that the function g = ∂f /∂η does not depend on ξ; therefore it is a function
solely of η, i.e.
∂f
= H(η) (33)
∂η
for some function H. We now solve this latter equation: f is the antiderivative of H(η)
[let us call this G(η)] plus a “constant of integration” that does not depend on η but
may depend on ξ [let us call this F (ξ)]. Therefore f (ξ, η) = G(η) + F (ξ).

Returning to the variables x and t, it follows that the general solution to the one-dimensional
wave equation (11) is
f (x, t) = F (x − ct) + G(x + ct) . (34)

3.3 Interpretation as the initial-value problem

We can understand the general solution (34) in another way, in connection with the
initial-value problem for the one-dimensional wave equation. Recall first that an nth-
order ordinary differential equation for a function f (t) has a general solution that involves
n arbitrary “constants of integration”; and these n numbers can be determined, for any
particular solution, by specifying n initial conditions (usually the value of f and its first
n − 1 derivatives at some initial time). Likewise, a partial differential equation that is of nth
order in time, for a function f (t, x) [here x denotes one or more spatial variables], will have
a general solution that involves n arbitrary “functions of integration” (that is, “constants of
integration” that are constant as concerns t but can be arbitrary functions of x); and these
n functions can be determined, for any particular solution, by specifying n initial functions
(usually the value of f and its first n − 1 partial derivatives with respect to time, throughout
space, at some initial time).

10
 Since the wave equation (11) [or (13)] is of second order in time, we will need to specify
two initial functions: for instance, the values of f and ∂f /∂t throughout space at time 0.
Let us denote these by ϕ and ψ:
def
ϕ(x) = f (x, 0) (35a)
def ∂f
ψ(x) = (x, 0) (35b)
∂t
In our general solution f (x, t) = F (x − ct) + G(x + ct), we have
ϕ(x) = F (x) + G(x) (36a)
ψ(x) = c [G′ (x) − F ′ (x)] (36b)
Letting Ψ(x) be an antiderivative of ψ(x), it follows that
Ψ(x)
G(x) − F (x) = +κ (37)
c
where κ is some constant. We then have
1h Ψ(x) i
F (x) = ϕ(x) − − κ (38a)
2 c
1h Ψ(x) i
G(x) = ϕ(x) + + κ (38b)
2 c
and hence
f (x, t) = F (x − ct) + G(x + ct) (39a)
ϕ(x − ct) + ϕ(x + ct) Ψ(x + ct) − Ψ(x − ct)
= + (39b)
2 2c
x+ct
ϕ(x − ct) + ϕ(x + ct) 1
Z
= + ψ(s) ds . (39c)
2 2c
x−ct

This explicit expression for the solution of the one-dimensional wave equation in terms of
the initial data is called d’Alembert’s formula.10

3.4 Normal modes via separation of variables

Let us now consider the normal modes of an infinite string: that is, we are interested
in the solutions of the one-dimensional wave equation (with x ranging over the entire real
line) that are of the special form
f (x, t) = g(x) eiωt . (40)
10
Jean-Baptiste le Rond d’Alembert (1717–1783) was a French mathematician and physicist who made
many important contributions (some of which we will see later in this course). He studied the one-dimensional
wave equation, and found this general formula for its solution, in a 1747 paper.

11
But it is instructive to see how the “oscillatory Ansatz” (40) arises as the outcome of a more
general technique called separation of variables: here we search for solutions of the more
general form
f (x, t) = g(x) h(t) (41)
without imposing any a priori requirement on what the function h (or g) is.
So let us plug the “separation-of-variables Ansatz” (41) into the one-dimensional wave
equation (11): we obtain
1
g ′′ (x) h(t) = 2 g(x) h′′ (t) (42)
c
or, rearranging,
g ′′ (x) h′′ (t)
c2 = . (43)
g(x) h(t)
Note now that the left-hand side of (43) depends only on x, while the right-hand side depends
only on t. Therefore, if they are equal to each other, they must both be equal to the same
constant, i.e. independent of both x and t! Let us call this constant κ. We then have the
pair of equations

g ′′ (x) κ
= 2 (44a)
g(x) c
h′′ (t)
= κ (44b)
h(t)

or equivalently

d2 g κ
− g = 0 (45a)
dx2 c2
d2 h
− κh = 0 (45b)
dt2
The general solutions are therefore
√ √
g(x) = Ae( κ/c)x
+ Be−( κ/c)x
(46a)
√ √
κt κt
h(t) = Ce + De− (46b)

If κ > 0 these are exponential functions, which get very large as x (or t) tends to +∞ or
−∞ or both, and are therefore not of much physical interest (we have assumed that the
vibrations of our string have small amplitude, and these solutions violate that assumption).
So let us concentrate on the case κ < 0: for convenience we write κ = −ω 2 , so that the
solutions are

g(x) = Aei(ω/c)x + Be−i(ω/c)x (47a)

h(t) = Ceiωt + De−iωt (47b)

12
We may then write the solution f (x, t) = g(x) h(t) in a shorthand manner as

f (x, t) ∼ e±i(ω/c)x e±iωt (48a)

∼ e±i(ω/c)(x±ct) (48b)

where what this means is that the actual f (x, t) arising from our separation-of-variables
Ansatz is a linear combination of the four expressions gotten from making all possible choices
of + and − signs in (48), namely

f (x, t) = A1 ei(ω/c)(x+ct) + A2 ei(ω/c)(x−ct) + A3 e−i(ω/c)(x+ct) + A4 e−i(ω/c)(x−ct) . (49)

These four contributions, taken for all possible ω, are the normal modes for the one-
dimensional wave equation in infinite space.

Another way of expressing this is to say that the normal modes are

f (x, t) = ei(kx±ωt) , (50)

where the wavenumber k ∈ R is related to the angular frequency ω ∈ [0, ∞) by

ω = c |k| , (51)

where c is the speed of propagation. Since the wavelength is λ = 2π/k and the
frequency is f = ω/2π, the relation (51) can be rewritten as

fλ = c, (52)

which you probably learned in high school.

For the n-dimensional wave equation (13) the normal modes are

f (x, t) = ei(k·x±ωt) (53)

where the wavenumber k ∈ Rn is related to the angular frequency ω ∈ [0, ∞) by

ω = c |k| . (54)

In general, formulae expressing the angular frequency ω as a function of the wavenum-

ber k are called dispersion relations. The dispersion relation for the (one-dimensional
or n-dimensional) wave equation is linear , in the sense that ω is strictly proportional
to |k|. By contrast, other linear partial differential equations with constant coefficients
can have nonlinear dispersion relations.

Note that in infinite space (unlike the case of a finite string tied down at the ends) there
are no boundary conditions forcing the frequency ω to lie in some special set of values; all
frequencies ω give rise to legitimate solutions.
The general solution of the one-dimensional wave equation in infinite space is then a
linear combination of these normal modes. Since ω ranges over the entire real line (rather
than just a discrete set), this “linear combination” may have to be interpreted as an integral
rather than as a finite or countably infinite sum. Since the normal modes are just complex
exponentials in both space and time, what we are really doing is Fourier analysis.

13
4 Solutions of the one-dimensional wave equation on a
finite interval
Let us now take a look at the solutions of the one-dimensional wave equation (11) when
the spatial variable x ranges over a finite interval of the real line, say 0 ≤ x ≤ L. In this case
the partial differential equation (11) has to be supplemented with boundary conditions
specifying how the solution f (x, t) is constrained to behave at the two endpoints x = 0 and
x = L. The two most commonly arising types of boundary conditions are:

• Dirichlet boundary conditions. We impose f = 0 at the given endpoint, at all

times. If f is the transverse displacement of a vibrating string, this means that the
string is tied down at the given endpoint. If f is the air pressure in an organ pipe, this
means that the pipe is open at the given endpoint.11

• Neumann boundary conditions. We impose ∂f /∂x = 0 at the given endpoint, at

all times. If f is the transverse displacement of a vibrating string, this means that at
the given endpoint we attach a loop to the string and allow it to slide frictionlessly
along a vertical pole.12 If f is the air pressure in an organ pipe, this means that the
pipe is closed at the given endpoint.

Let us not attempt to find the general solution to the one-dimensional wave equation on
a finite interval with boundary conditions, but simply pass to the computation of the normal
modes.

4.1 Normal modes via separation of variables

We can copy verbatim the analysis from Section 3.4, up through equation (47): we make
the separation-of-variables Ansatz f (x, t) = h(x) g(t) and find the solution

g(x) = Aei(ω/c)x + Be−i(ω/c)x (55a)

h(t) = Ceiωt + De−iωt (55b)

or equivalently

g(x) = A′ cos(ωx/c) + B ′ sin(ωx/c) (56a)

h(t) = C ′ cos(ωt) + D ′ sin(ωt) (56b)

Now we impose the boundary conditions, first at x = 0:

11
At the end of a pipe open to the air, the pressure cannot oscillate; rather, it is fixed at the ambient
pressure p0 of the surrounding air. Since the constant function p0 solves the one-dimensional wave equation,
and this equation is linear, we can make a change of dependent variable f → f −p0 , i.e. we can reinterpret f as
meaning the difference between the observed pressure and the ambient pressure. With this reinterpretation,
we have the boundary condition f = 0 at an open end of the pipe.
12
The loop is assumed massless. Therefore the net vertical force on it must be zero, so the tangent to the
string must be horizontal at the endpoint, i.e. ∂f /∂x = 0.

14
 • If the boundary condition at x = 0 is Dirichlet [i.e., f (0, t) = 0 for all t], then we must
choose g(0) = 0, hence A′ = 0 — that is, the g solution is pure sine.
• If the boundary condition at x = 0 is Neumann [i.e., (∂f /∂x)(0, t) = 0 for all t], then
we must choose g ′(0) = 0, hence B ′ = 0 — that is, the g solution is pure cosine.
We then impose in the same way the boundary condition at x = L, namely g(L) = 0 for
a Dirichlet boundary condition or g ′ (L) = 0 for a Neumann boundary condition; this will
restrict the frequency ω to lie in a suitable discrete set. There are four cases:
• Dirichlet at both ends. The Dirichlet condition at x = 0 forces g(x) ∼ sin(ωx/c).
The Dirichlet condition at x = L then forces ωL/c to be an integer multiple of π, i.e.
πc
ω = j for j = 1, 2, . . . . (57)
L

• Dirichlet at one end and Neumann at the other. There are two possibilities
(D–N and N–D): let us consider the first. The Dirichlet condition at x = 0 again
forces g(x) ∼ sin(ωx/c). The Neumann condition at x = L then forces ωL/c to be an
odd-integer multiple of π/2, i.e.
πc πc
ω = (2j − 1) = (j − 12 ) for j = 1, 2, . . . . (58)
2L L

The second possibility (N–D) can be treated analogously. But there is no need: the
two possibilities (D–N and N–D) are interchanged by x ↔ L − x (which leaves the
wave equation invariant), so the frequencies are the same for both (and the normal
modes are related by x ↔ L − x).

• Neumann at both ends. The Neumann condition at x = 0 forces g(x) ∼ cos(ωx/c).

The Neumann condition at x = L then forces ωL/c to be an integer multiple of π, i.e.
πc
ω = j for j = 1, 2, . . . . (59)
L
(In principle j = 0 is also allowed, but this corresponds to the trivial solution f (x, t) =
const of the one-dimensional wave equation.) The frequencies are thus the same as in
the Dirichlet-at-both-ends case, although the normal modes look very different (mul-
tiples of half-cycles of cos instead of sin).
Thus, for instance, an organ pipe that is closed at one end and open at the other will
have a fundamental frequency (j = 1) that is half that of a pipe of the same length that is
open at both ends, i.e. it will sound one octave lower. Also, a pipe that is closed at one end
and open at the other resonates only at odd harmonics, while a pipe that is open at both
ends resonates at all harmonics.

We can use these normal modes to solve the initial-value problem: the idea is to
determine the coefficient of each normal mode by Fourier-analyzing the initial data ϕ(x) =
f (x, 0) and ψ(x) = ∂f
∂t
(x, 0). But we have probably already done enough for an introduction
to waves!

15