Solutions to the wave equation

September 16, 2019

We have derived the 2-dimensional wave equation

1 ∂2q ∂2q
− + =0 (1)
c2 ∂t2 ∂x2
in two different ways as a continuum limit from Newton’s second law. Just as generic potentials give rise to
simple harmonic oscillation at any minimum of the potential, extended bodies almost always have wavelike
solutions. We now explore some solutions to Eq.(1).

1 Standing wave solution

Recall that in solving for N coupled oscillators, we assumed that the displacements of the N masses varied
sinusoidally and that the time dependence was oscillatory, eiΩt . Let’s try the same sort of solution for the
wave equation. Let
q (x, t) = A sin ωt sin kx
for some constants ω, k. Substituting into Eq.(1),

1 ∂2 ∂2
− (A sin ωt sin kx) + (A sin ωt sin kx) = 0
c2 ∂t2 ∂x2
1
− 2 −ω 2 (A sin ωt sin kx) − k 2 (A sin ωt sin kx) = 0


c
Collecting terms

ω2
 
2
− k A sin ωt sin kx = 0
c2

we see that we have a solution as long as ω and k are related by

ω
k=±
c
Notice that ω has units of frequency and k, called the wave number, has units of inverse length. The
amplitude, A, and the frequency, ω, are arbitrary, so we may write

q (x, t) = A sin ωt sin kx

ωx
= A sin ωt sin
c
for any ω, A.

1
1.1 Boundary conditions and normal modes
In deriving Eq.(1) we let the number of oscillators become infinite, N → ∞. Since the number of normal
modes of an N oscillator system is N , we might guess that there will now be infinitely many normal modes.
This is exactly right.
To find the normal modes, we must impose boundary conditions. Let the wave be a guitar string of
length L, fixed at both ends. Measuring x from one end, we therefore require at x = 0 and x = L

q (0, t) = 0
q (L, t) = 0

The first of these is automatically satisfied, q (0, t) = A sin ωt sin 0 = 0. For the second

0 = q (L, t)
= A sin ωt sin kL

Since this must hold at all times t, we demand

sin kL = 0
ωL
so that c must be a multiple of π,

kL = nπ
nπ
k =
L
and for the possible frequencies,
nπc
ω =
L
The normal modes of the string are therefore

q (x, t) = A sin ωt sin kx

 
nπct  nπx 
= A sin sin
L L
There is a simple way to think of this. The x-dependent terms gives the shape of an envelope of the
motion, while the time dependence produces oscillation within this spatial limit. For example, when n = 1
the spatial dependence is  πx 
A sin
L
As x runs from 0 to L, A sin πx 1


L runs through 2 cycle, from a zero
through a single maximum then back
to zero–a half wavelength. The remaining time dependence sin nπct L causes movement of the string up and
down within the limits of this half wave.
Changing n increases the number of nodes. For n = 2 the envelope comprises an entire wavelength, for
n = 3 there are two nodes and 1 12 wavelengths. In general, the envelope includes n2 wavelengths, so the
wavelength is λ = 2L
n . Therefore,

2L
λ =
n
2π
=
k
nπ
since k = L . Thus, the wave number k is related to the wavelength by
2π
k=
λ

2
 If instead, we look at a fixed position and ask for the time between successive peaks, we find the period,
T , satisfies

ωT = 2π
2π
ω =
T

1.2 Traveling waves

A standing wave may be understood as a superposition of two traveling waves moving in opposite directions.
Suppose we write another solution to the wave equation,

q (x, t) = Re Aeiωt cos kx


(2)

with A complex. Letting A = 2A0 e−iϕ0 so that

n o
q (x, t) = Re 2A0 ei(ωt−ϕ0 ) cos kx

and the real part is just

q (x, t) = 2A0 cos (ωt − ϕ0 ) cos kx

For simplicity, choose zero phase at the initial time, t0 = 0, so that
ϕ0 = 0. This is another standing wave
with a shifted phase for the spatial part (since cos kx = sin kx + π2 ).
However, suppose we start again with Eq.(2) and write the cosine as
1 ikx
e + e−ikx


cos kx =
2
Then (still setting ϕ0 = 0 for simplicity) q (x, t) becomes
  
iωt 1 ikx −ikx


q (x, t) = Re 2A0 e e +e
2
= Re A0 eiωt eikx + e−ikx


n  o
= Re A0 ei(ωt+kx) + ei(ωt−kx)
= A0 cos (ωt + kx) + A0 cos (ωt − kx)

First, we check that A0 cos (ωt ± kx) satisfies the wave equation,

1 ∂2q ∂2q A0 ∂ 2 ∂2
− 2 2
+ = − (cos (ωt ± kx)) + A0 2 (cos (ωt ± kx))
c ∂t ∂x2 2
c ∂t 2 ∂x
A0 2


2

= − 2 −ω cos (ωt ± kx) + A0 − (±k) cos (ωt ± kx)
 c2 
ω 2
= − k A0 cos (ωt ± kx)
c2
ω
so once again we have a solution as long as k = c.
The solution

q (x, t) = A0 cos (ωt + kx) + A0 cos (ωt − kx)

is a superposition of a right moving wave and a left moving wave. We can see this as follows.

3
 How do these waves appear to move? Since a maximum of the cosine occurs when its argument is a
multiple of 2π, the maximum of the first term occurs at any x and t such that

ωt + kx = 2nπ

Solving for x,
2nπ ω
xL
max = − t
k k
we see that the position of any one maximum of the first cosine moves with velocity

dxL
max ω
= − = −c
dt k
For the second cosine, the position of any given maximum moves with velocity

dxR
max
= +c
dt
The solution therefore represents a superposition of a wave with crests moving to the left and a wave with
crests moving to the right.
We have shown that a standing wave may also be seen as a superposition of a right moving wave and a
left moving wave.

2 General solutions
2.1 Superposition
Because the wave equation is linear, we may take arbitrary linear combinations of solutions. Most generally,
we may sum solutions with any frequency ω, and the amplitude A may be different for each different ω. We
may add together n solutions with frequencies ωm , m = 1, . . . , n,
n
X ωm x
q (x, t) = A (ωm ) eiωm t sin
m=1
c

While this can give a wide variety of waveforms, we can write an even more general wave by taking a
superposition of all values of ω and integrating. Let A (ω) be the average amplitude per unit frequency in
an infinitisimal range dω. Then
ˆ∞
ωx
q (x, t) = A (ω) eiωt sin dω
c
0

is still a solution to the wave equation. The amplitude density can give the wave any shape we like.

2.2 Integration
We can sometimes directly integrate partial differential equations. The 2-dimensional wave equation takes a
simpler form if we change variables. Let

u = x − ct
v = x + ct

4
We use the chain rule to transform the derivatives of any function q (x, t):
∂q ∂u ∂q ∂v ∂q
= +
∂x ∂x ∂u ∂x ∂v
∂q ∂q
= +
∂u ∂v
∂q ∂u ∂q ∂v ∂q
= +
∂t ∂t ∂u ∂t ∂v
∂q ∂q
= −c +c
∂u ∂v
Since this is true for any function q, we may write it as a relation between the derivative operators,
∂ ∂u ∂ ∂v ∂
= +
∂x ∂x ∂u ∂x ∂v
∂ ∂
= +
∂u ∂v
∂ ∂u ∂ ∂v ∂
= +
∂t ∂t ∂u ∂t ∂v
∂ ∂
= −c +c
∂u ∂v
Therefore, in terms of u and v the wave equation becomes
1 ∂2q ∂2q
0 = − 2 2
+
c  ∂t ∂x2     
1 ∂ ∂ ∂q ∂q ∂ ∂ ∂q ∂q
= − 2 −c +c −c +c + + +
c ∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v
 2
∂2q ∂2q ∂2q ∂2q ∂2q ∂2q ∂2q
  
∂ q
= − − − + + + + +
∂u2 ∂v∂u ∂u∂v ∂v 2 ∂u2 ∂v∂u ∂u∂v ∂v 2
∂2q
= 4
∂v∂u
We may divide away the 4 and look at the resulting equation as
 
∂ ∂q
=0
∂v ∂u
∂q
This shows that ∂u is independent of v so we may write
∂q
= F (u)
∂u
for an arbitrary function F (u). We may now integrate over u,
ˆ
q (u, v) = F (u) du + g

but we had only a partial derivative. This means that the “constant” we add may be a function of v:
ˆ
q (u, v) = F (u) du + g (v)

∂q
Check that this does satisfy ∂u = F (u), since the partial derivative of g (v) with respect to u is zero. Finally,
we may rewrite the indefinite integral over u as a new function of u,
ˆ
f (u) := F (u) du

5
so that the full solution is
q (u, v) = f (u) + g (v)
with the functions f and g arbitrary. Substituting back,

q (x, t) = f (x − ct) + g (x + ct) (3)

This form has a clear physical interpretation for a wave on a string. At t = 0 the disturbance of the
string is given by f (x) + g (x); as time progresses, two waveforms move apart. One has the same shape as
f (x) but moves to the right at speed c, while the other maintains the shape g (x) but moves to the left with
speed c.
Either f (x − ct) alone or g (x + ct) alone still solves the wave equation. This means that if we choose
a solution with f (x − ct) alone, it describes a wave with an arbitrary shape f (x) when t = 0. This shape
is preserved the wave moves, but the shape moves off to the right with speed c. Similarly, g (x + ct) alone
describes an initial shape g (x) that moves to the left without distorting.

2.3 Initial value solution and d’Alembert’s formula

Our solution for a wave on a string becomes unique when we impose boundary conditions such as holding
the endpoints of the string fixed. Another common form for specifying a unique solution–as you have seen
in solving Newton’s second law–is the initial value formulation. For Newton’s second law, specifying initial
values amounts to giving the initial position and velocity vectors for each particle in the system. When we
change to a continuum equation, this means we need a continuum of initial values.
Looking at our general solution, Eq.(3) we see that at time t = 0 we still have

q (x, 0) = f (x) + g (x)

where f (x) and g (x) remain arbitrary. This sum corresponds to the initial displacement of every particle
of the system (string, slinky, organ pipe, etc.). We may also ask for the initial velocity,

df ∂u df ∂v
q̇ (x, 0) = +
du t=0 ∂t dv t=0 ∂t
= −cf 0 (x) + cg 0 (x)

We now show that specifying these two functions gives a unique solution via a general formula due to
d’Alembert.
Let

q (x, 0) = : a (x)
q̇ (x, 0) = : b (x)

be given functions. Then, writing equations for f (x) and g (x), we have

f (x) + g (x) = a (x)

0 0
c (−f (x) + g (x)) = b (x)

where f 0 and g 0 denote ordinary derivatives of f and g with respect to their arguments, then evaluated at
u (x, 0) = x and v (x, 0) = x. We may rewrite the second equation using an arbitrary variable y as
 
df dg
c − (y) + (y) = b (y)
dy dy

6
Multiply by dy and integrate from any initial value x0 to x,
 
df dg
c − (y) + (y) dy = b (y) dy
dy dy
ˆx ˆx ˆx
1
− df (y) + dg (y) = b (y) dy
c
x0 x0 x0
ˆx
1
−f (x) + f (x0 ) + g (x) − g (x0 ) = b (y) dy
c
x0
ˆx
1
−f (x) + g (x) = b (y) dy − f (x0 ) + g (x0 )
c
x0

Now we can solve,

f (x) + g (x)
= a (x)
ˆx
1
−f (x) + g (x) = b (y) dy − f (x0 ) + g (x0 )
c
x0
ˆx
 
1 1
g (x) = a (x) + b (y) dy − f (x0 ) + g (x0 )
2 c
x0
ˆx
 
1 1
f (x) = a (x) − b (y) dy + f (x0 ) − g (x0 )
2 c
x0

where
g (x0 ) + f (x0 ) = a (x0 )
The full solution for the displacement at all times is therefore,
q (x, t) = f (x − ct) + g (x + ct)
ˆ ˆ
 x−ct
  x+ct

1 1 1 1
= a (x − ct) − b (y) dy + f (x0 ) − g (x0 ) + a (x + ct) + b (y) dy − f (x0 ) + g (x0 )
2 c 2 c
x0 x0
ˆ ˆ
 x−ct x+ct

1 1 1
= a (x − ct) + a (x + ct) − b (y) dy + f (x0 ) − g (x0 ) + b (y) dy − f (x0 ) + g (x0 )
2 c c
x0 x0
ˆ ˆ
 x−ct x+ct

1 1 1
= a (x − ct) + a (x + ct) − b (y) dy + b (y) dy 
2 c c
x0 x0

The difference of the two integrals is

ˆ
x−ct ˆ
x+ct ˆx0 ˆ
x+ct
1 1 1 1
− b (y) dy + b (y) dy = b (y) dy + b (y) dy
c c c c
x0 x0 x−ct x0
ˆ
x+ct
1
= b (y) dy
c
x−ct

7
so the final expression for q (x, t) is

ˆ
 x+ct

1 1
q (x, t) = a (x − ct) + a (x + ct) + b (y) dy 
2 c
x−ct

This is d’Alembert’s formula. Notice that it depends only on the two functions a (x) and b (x) with all
reference to x0 dropping out.
It is now clear that

3 Gaussian wave packets

An important class of solutions is described by a Gaussian or normal distribution:
x2
a (x) = Ae− 2σ2

This function has a maximum at x = 0 and falls off smoothly to zero as x approaches infinity. The √ speed of
the fall-off is determined by the constant σ (called the standard deviation in statistics). When x = 2σ the
value has dropped to 1e = .37... of its original value A.

3.1 Normalizing
Often, the constant A is chosen so that the area under the Gaussian is one,
ˆ∞
x2
1=A e− 2σ2 dx
−∞

There is a clever trick for evaluating this integral. Let

x2
I = e− 2σ2

and consider I 2 , written as two independent integrals,

ˆ∞ ˆ∞
x2 y2
−
I 2
= e 2σ 2 dx e− 2σ2 dy
−∞ −∞
ˆ∞ ˆ∞ (
x2 +y 2 )
= e− 2σ 2 dxdy
−∞ −∞

If we think of this as an integration over the whole xy plane, we may rewrite it in polar coordinates. Since
r2 = x2 + y 2 and an area element in polar coordinates is rdrdϕ, we may change coordinates and write

ˆ∞ ˆ2π
r2
2 −
I = e 2σ 2 rdr dϕ
0 0
ˆ∞ 2
− r2
= 2π e 2σ rdr
0

8
where the angular integral is trivial. A simple substitution, w = r2 with dw = 2rdr, lets us perform the
second integral,
ˆ∞ ˆ∞  
− r2 w 1
e 2σ 2 rdr = e− 2σ2 dw
2
0 0
ˆ∞
1 w
= e− 2σ2 dw
2
0
w ∞
= −σ 2 e− 2σ2
0
∞
= −σ 2 e− 2σ2 + σ 2 e−0
= σ2

Therefore,

I2 = 2πσ 2
√
I = 2πσ 2

Choosing A = √ 1 gives us the result

2πσ 2

ˆ∞
1 x2
√ e− 2σ2 dx = 1
2πσ 2
−∞

This normalized Gaussian is used in probability and in quantum mechanics. We will use it soon to give
a description of the Dirac delta function.

3.2 Gaussian waves

Choose a wave with the initial conditions
x2
a (x) = Ae− 2σ2
b (x) = 0

Then we have the immediate solution

 
A (x−ct)2 (x+ct)2
q (x, t) = e− 2σ2 + e− 2σ2
2

This describes a pair of Gaussian shapes of half the amplitude, moving off in opposite directions.