Problem Solving in Theoretical Physics

Fourier and Laplace Transforms – Useful Results/Formulae

1 Fourier Transforms

1.1 Fourier Series

We begin with a brief revision of the more familiar Fourier series. We may represent a
function defined on some finite interval, [−L, L], by an infinite sum of plane waves
∞
X inπx
f (x) = Cn exp . (1)
n=−∞
L

To calculate the coefficients Cn , we use

Z π Z L  
dk dy iπ(m − n)y kL
δmn = exp (ik(m − n)) = exp y= . (2)
−π 2π −L 2L L π

Then
L   Z L ∞  
iπ(m − n)x
Z
dx iπnx dx X
f (x) exp − = Cm exp
−L 2L L −L 2L m=−∞ L
∞ (3)
X
= Cm δmn = Cn .
m=−∞

We can also construct Fourier sine and cosine series which involve either sines and cosines
in place of the exponential and therefore have a definite symmetry about the origin. If a
function f (x) is defined on [0, L], we are free to choose either the odd or even extension
to [−L, L]. If f (0) 6= 0, it is more natural to choose the cosine (even) extension to avoid
a discontinuity.
From Eq. (2), it is easy to show that
Z L
dx  πmx   πnx 
sin sin = δmn and
−L L L L
Z L (4)
dx  πmx   πnx 
cos cos = δmn for m, n > 0
−L L L L

1
 Problem Solving in Theoretical Physics

(or you can use the angle-addition formulae). It is therefore straight-forward to invert
a sine or cosine series, for example, given:
∞
X  nπx 
f (x) = An sin ,
n=1
L
(5)
Z L Z L
1  mπx  2  mπx 
Am = f (x) sin dx = f (x) sin dx.
L −L L L 0 L

1.2 Transform Definition and Inverse

We define the Fourier Transform (FT) of a function f (x) as
Z ∞
F[f (x)] = dx f (x)e−ikx . (6)
−∞

It is often denoted g(k) or f˜(k). This is well-defined for functions where f (x) → 0 as
x → ±∞ so that the integral converges. The inverse is given by
Z ∞
−1 1
f (x) = F [g(k)] = dk g(k)eikx . (7)
2π −∞

This can be verifed by using

Z ∞
dk ikx
δ(x) = e . (8)
−∞ 2π

The FT may easily be generalised to higher dimensions, for example in three dimensions
Z ∞ Z ∞
3 −ik·x 1
F[f (x)] = g(k) = d x f (x)e and f (x) = 3
d3 k g(k)eik·x (9)
−∞ (2π) −∞

where k is a point in 3-d space.

1.3 General Properties

The Fourier Transform of a derivative:
Z ∞ Z ∞
0 0 −ikx −ikx ∞
dx f (x)(−ik)e−ikx
 
F[f (x)] = dx f (x)e = f (x)e −∞
−
−∞ −∞ (10)
= ikF[f (x)].

Similarly

F[f 00 (x)] = −k 2 F[f (x)], ..., F[f (n) (x)] = (ik)n F[f (x)]. (11)

2
 Problem Solving in Theoretical Physics

The convolution of two functions f1 (x) and f2 (x) is defined as

Z ∞
(f1 ∗ f2 )(x) = f1 (y)f2 (x − y)dy . (12)
−∞

The idea is that we are integrating up the product of f1 and f2 where the sum of the
arguments is constant. This can arise for example as the probability that the sum of
two independent random variables with probabilities f1 and f2 takes value x.
Now,
Z ∞ Z ∞
−ikx
F[f1 ∗ f2 ] = dxe dy f1 (y)f2 (x − y)
Z−∞
∞ Z−∞∞
0
= dy f1 (y) dx0 e−ik(x +y) f2 (x0 )
−∞ −∞ (13)
Z ∞ Z ∞
−iky
= dy f1 (y)e dx f2 (x)e−ikx
−∞ −∞
= F[f1 (x)]F[f2 (x)].

Thus the Fourier Transform of a convolution factorises into a product of the Fourier
Transform of each function. This can be very useful when we come to invert the trans-
form.
Importantly there is a converse to this:
Z ∞
1 1
F[f1 (x)f2 (x)] = g1 (k 0 )g2 (k − k 0 )dk 0 = (g1 ∗ g2 )(k), (14)
2π −∞ 2π

the Fourier Transform of the product of two functions is equal (up to a multiplicative
constant) to the convolution of the two Fourier Transforms.

1.4 ODEs with Fourier Transforms

We can exploit the simplicity of the Fourier Transform of a derivative to solve ODEs.
We frequently need to apply the convolution theorem too. This is best illustrated in an
example. Consider Forced Damped Harmonic Motion, where we wish to solve:

ÿ(t) + 2κẏ + Ω2 y(t) = f (t). (15)

By convention, since t is the independent variable we use ω as the Fourier variable

instead of x. We let F[y(t)] = ye(ω) and F[f (t)] = fe(ω). Then the Fourier Transform
of Eq. (15) gives

(−ω 2 + 2iκω + Ω2 )e
y (ω) = fe(ω)
1 (16)
⇒ ye(ω) = fe(ω).
Ω + 2iκω − ω 2
2

3
 Problem Solving in Theoretical Physics

To invert this, we note that this can be viewed as a product of two Fourier Transforms
and we may therefore invoke the convolution theorem, Eq. (13), to get
Z ∞
y(t) = dt0 G(t − t0 )f (t0 )
−∞
 
−1 1
where G(t) = F (17)
Ω2 + 2iκω − ω 2
Z ∞
dω eiωt √
=− with ω± = iκ ± Ω2 − κ2 .
−∞ 2π (ω − ω+ )(ω − ω− )

We recognise that G(t) is the Green function of the problem and we will assume that
it vanishes as t → ±∞ so that its Fourier Transform exists. We will discuss this further
in the next subsection. In the meantime, we evaluate the inversion integral for G(t) by
closing the contour and using the residue theorem.
For t > 0 we close the contour in the upper half plane (because we wish to apply
Jordan’s Lemma and the integrand has a factor of e+iωt ). In this case G(t) depends on
the position of the poles ω± because their imaginary parts are positive for all Ω and κ.

eiωt
 
X 1
G(t) = −2πi Res
ω±
2π (ω − ω+ )(ω − ω− )
(18)
eiω+ t eiω− t
 
= −i + .
(ω+ − ω− ) (ω− − ω+ )

For t < 0 we close the contour in the lower half plane (because of the e−iω|t| ) and the
contour does not contain any residues. Thus G(t) = 0 for t < 0.
We interpret the equation
Z ∞
yp (t) = dt0 G(t − t0 )f (t0 ) (19)
−∞

as the integral of the stimulus f at time t0 × the response G(t − t0 ) at time t. In the
example G(t − t0 ) = 0 for t < t0 i.e. there is no response in the past (t) to a stimulus in
the future (t0 ) and so the system is causal.

2 Laplace Transforms

2.1 Definition and Inverse

The Laplace transform of a function f (t) is defined to be
Z ∞
L[f (t)] = F (s) = dt f (t)e−st . (20)
0

4
 Problem Solving in Theoretical Physics

This will only converge for <[s] sufficiently large. For example,
Z ∞
1
ωt
L[e ] = dt e−(s−ω)t = (s > ω). (21)
0 s−ω
For most functions, a sufficiently large s can be found. There are a few exceptions, for
2
example the Laplace Transform of et does not exist. There may also be problems with
convergence at t = 0; the Laplace Transform of tn does not exist for n ≤ −1.
The restriction to positive t is necessary for convergence. However, in many dynamical
problems we are only interested in the behaviour for t > 0 subject to some initial
conditions at t = 0.
The inverse transform is given by the Bromwich integral in the complex plane
Z γ+i∞
1
f (t) = ds F (s)est . (22)
2πi γ−i∞

The constant γ must be chosen to be to the right of any singularity of F (s) in the
complex s plane, and be consistent with the condition on <[s]. For t > 0 the contour
can be closed as a large semicircle to the left (which vanishes by a Jordan’s lemma-type
argument), see Fig. 1. One can then use the residue theorem to obtain f (t) from F (s).

s
s3

s1

s2

Figure 1: The blue dashed line shows the contour for the Bromwich inversion integral
for a function F (s) with singularities at s1 , s2 and s3 . γ must be chosen to the right of
the singularities. For t > 0 we close contour to the left, shown by the black dashed line.

Example 1: F (s) = 1/s

We have a simple pole at s = 0 and thus for t > 0,

f (t) = Residue of est /s = 1 (23)

so f (t) = θ(t).

5
 Problem Solving in Theoretical Physics

Example 2: F (s) = 1/sm+1 for m integer.

We have a pole of order m + 1 at s = 0 and thus for t > 0,
tm
f (t) = Residue of est /sm+1 = , (24)
m!
tm
and so f (t) = m!
θ(t). This agrees with L[tν ] = Γ(ν + 1)/sν+1 .

2.2 Common Examples and Properties

The following functions commonly occur and are listed below with their Laplace Trans-
forms. These may be easily checked.
R∞ R ∞ ν −u
f (t) = tν F (s) = 0 tν e−st dt = sν+1
1
0
u e du = Γ(ν+1)
sν+1
(s > 0)
1
f (t) = eωt F (s) = s−ω
(s > ω)
1 s+iω
f (t) = eiωt F (s) = s−iω
= s2 +ω 2
(s > 0)
1 s
f (t) = cos ωt F (s) = 2(s2 +ω 2 )
[s + iω + s − iω] = s2 +ω 2
(s > 0)
w
f (t) = sin ωt F (s) = s2 +ω 2
(s > 0)

f (t) = δ(t − a) F (s) = e−as (s > 0)

e−as
f (t) = θ(t − a) F (s) = s
(s > 0)

Similarly to the Fourier Transform, the Laplace Transform of a derivative has a simple
form:
Z ∞ Z ∞
0 0 −st −st ∞
f (t)e−st dt = sF (s) − f (0)
 
L[f (t)] = f (t)e dt = f (t)e 0
+s
0 0 (25)
L[f 00 (t)] =s2 F (s) − sf (0) − f 0 (0).

There is also a convenient expression for the derivative of the Laplace Transform itself:
Z ∞
d
L[f (t)] = − tf (t)e−st dt = −L[tf (t)]. (26)
ds 0

These properties will prove useful when we come to apply Laplace Transforms to ODEs.
We also have an analogous result for convolutions. In the context of Laplace transforms
we define the convolution of two functions f (t) and g(t) as
Z t
f (t) ◦ g(t) = f (t − z)g(z)dz. (27)
0

6
 Problem Solving in Theoretical Physics

The restricted limits come from the requirement that the arguments of the two functions
are positive. Taking the Laplace Transform gives
Z ∞ Z t
−st
L[f (t) ◦ g(t)] = dt e dz f (t − z)g(z)
Z0 ∞ Z ∞ 0
= dz dt e−st f (t − z)g(z)
(28)
Z0 ∞ Zz ∞
= dz du e−s(u+z) f (u)g(z) where u = t − z
0 0
= L[f ]L[g].

One can also show that

Z γ+i∞
1
L[f1 f2 ] = F1 (z)F2 (s − z)dz (29)
2πi γ−i∞

where F1 and F2 are the Laplace transforms of f1 and f2 respectively. If F1 (s) and F2 (s)
only exist for <[s] > α1 and <[s] > α2 respectively, we need <[s] − α2 > γ > α1 .

2.3 ODE Example

We again take damped harmonic motion as our example:

ü(t) + 2κu̇(t) + ω02 u(t) = 0, (30)

with initial conditions, u(0) = u0 and u̇(0) = v0 . Taking the Laplace transform gives

s2 F (s) − su(0) − u̇(0) + 2κ[sF (s) − u(0)] + ω02 F (s) = 0

(s + 2κ)u0 + v0 (31)
⇒ F (s) = where ω 2 ≡ ω02 − κ2 .
(s + κ)2 + ω 2

Now we have to invert F (s) to find u(t). We can use the inversion integral Eq. (22) as
follows. We know
Z γ+i∞
1 [(s + 2κ)u0 + v0 ] st
u(t) = ds e , s± = −κ ± iω. (32)
2πi γ−i∞ (s − s+ )(s − s− )

For t > 0, we close the contour with a large semicircle to the left which encloses the two
simple poles at s = s± . The sum of residues therefore gives

[(s+ + 2κ)u0 + v0 ] s+ t [(s− + 2κ)u0 + v0 ] s− t

u(t) = e + e (33)
(s+ − s− ) (s− − s+ )
e−κt
= [(κuo + v0 ) sin ωt + ωu0 cos ωt] . (34)
ω

7
 Problem Solving in Theoretical Physics

Now consider adding a forcing term to the right-hand side and setting the boundary
conditions to zero:

ü(t) + 2κu̇(t) + ω02 u(t) = f (t) u(0) = 0 u̇(0) = 0. (35)

Then denote the Laplace Transform of x(t), f (t) by x

e(s), fe(s). We find

fe(s)
x
e(s) =
(s + κ)2 + ω 2
Z t
⇒ x= G(t − t0 )f (t0 )dt0 (36)
0
Z γ+i∞
1 esu 1
where G(u) = ds 2 2
= e−κu sin ωu Θ(u).
2πi γ−i∞ (s + κ) + ω ω