CUST 2023-2024

Maths for Physics 2

Session 4

Fourier transform
Summary

Here we introduce the important notion of the Fourier transform, which can be
viewed as the generalization of Fourier series to nonperiodic functions on R. Indeed,
we discuss (heuristically) how the Fourier transform can be constructed from the
Fourier series of a periodic function in the limit of an infinite period. We then give
the precise mathematical definition of the Fourier transform, and discuss some of
the conventions that are typically used in the scientific literature. Finally, we derive
various properties of the Fourier transform.

i
Contents

1 General idea: from Fourier series to Fourier transforms 1

1.1 Fourier series of nonperiodic functions on a finite interval: reminders . 1
1.2 From Fourier series to Fourier transforms . . . . . . . . . . . . . . . . 2
1.3 Heuristic construction of the Fourier transform . . . . . . . . . . . . . 3

2 Mathematical definition of the Fourier transform 6

2.1 Frequent conventions for the Fourier transform . . . . . . . . . . . . . 6
2.2 The convention followed in this course . . . . . . . . . . . . . . . . . 8

3 Some properties of the Fourier transform 9

3.1 Linearity of the Fourier transform . . . . . . . . . . . . . . . . . . . . 9
3.2 Fourier transform of the complex conjugate . . . . . . . . . . . . . . . 10
3.3 Fourier transform of f (−t) . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4 Fourier transform of f (t − a) . . . . . . . . . . . . . . . . . . . . . . . 11
3.5 Fourier transform of f (at) . . . . . . . . . . . . . . . . . . . . . . . . 12

ii
Chapter 1

General idea: from Fourier series to

Fourier transforms

REMARK ON NOTATION: throughout these notes we will use the abbreviations

FT for Fourier Transforms and FS for Fourier Series.

1.1 Fourier series of nonperiodic functions on a fi-

nite interval: reminders
So far we discussed the notion of FS: we saw that a FS allows to represent a function
f (t) by a sum of cosines and sines (for the real FS) or, alternatively, by a sum of
complex exponentials (for the complex FS). But remember that such a representation
required the function f to be periodic: so what did we do for functions that were
not periodic?
Well, we kind of cheated: for a function that is not periodic on R (such as for
instance a polynomial function, e.g. f (t) = t), we only considered it on a finite
interval (t0 , t1 ), t0 < t1 , and then constructed the corresponding function fper (t),
such that

fper (t) = f (t) , ∀t ∈ (t0 , t1 ) , (1.1)

which then periodically repeats itself outside of (t0 , t1 ). Therefore, this function
fper (t), because it is by construction periodic, can be written as a FS fFS (t), i.e.

fper (t) = fFS (t) .

This was how we constructed the FS of the nonperiodic function f (t), that is

f (t) = fFS (t) , ∀t ∈ (t0 , t1 ) . (1.2)

1
CHAPTER 1. GENERAL IDEA: FROM FOURIER SERIES. . . 2

Such a representation of f (t) is of course only valid on the finite interval (t0 , t1 ):
outside of this interval, the two functions f (t) and fFS (t) are completely different.
Since this situation is not fully satisfactory, we now ask the question: what if
we want to go one step further, and construct a FS, or something analog, for a
nonperiodic function on an infinite interval, i.e. on the whole R? We now discuss
how we can actually do this, at the price of constructing a new mathematical object:
the so-called Fourier transform.

1.2 From Fourier series to Fourier transforms

Let’s start again from our idea of constructing a FS of a nonperiodic function f by
constructing a periodic function fper upon restricting f to a finite interval, which
we here take to be (−T /2, T /2) for some T > 0. That is, we require
 
T T
fper (t) = f (t) , ∀t ∈ − , , (1.3)
2 2

so that we can write fper as a FS, and thus f itself as a FS fFS on (−T /2, T /2), i.e.
 
T T
f (t) = fFS (t) , ∀t ∈ − , . (1.4)
2 2

It will prove convenient to our subsequent analysis to use the complex FS: therefore,
we write (1.4) as
∞  
X
2niπt/T T T
f (t) = cn e , ∀t ∈ − , , (1.5)
n=−∞
2 2

where the coefficients cn are thus given by

 T /2
1
cn = dt e−2niπt/T f (t) , ∀n ∈ Z . (1.6)
T −T /2

Now, let’s recall that our choice of the quantity T (which defines the period of
the periodic function fper (t) that we construct from f ) is completely arbitrary: we
can indeed choose whatever value we want for T . This is actually interesting for our
present purposes, because it means that we can in particular take T to be actually
as large as we want! Clearly, by taking larger and larger values of T , the periodic
function fper (t) matches the nonperiodic function f (t) over broader and broader
intervals, and hence becomes more and more representative of our actual function
f . And so here we reach the following important observation: if we consider the
limit where T → ∞, we can actually expect our periodic function fper to be basically
indistinguishable of our nonperiodic function f !
CHAPTER 1. GENERAL IDEA: FROM FOURIER SERIES. . . 3

This is indeed the key idea that will allow us to construct the so-called Fourier
transform, which can be intuitively understood as a kind of limit of the FS for a
periodic function whose period T goes to infinity.

1.3 Heuristic construction of the Fourier transform

DISCLAIMER: here we do not claim to be rigorous! Our aim is only to give some
idea of how we can reach, in a more or less natural way, the mathematical definition
of the FT that will be discussed in chapter 2 below.

To make the above idea a bit more quantitative, and obtain the mathematical
expression of the FT, let’s consider again the complex FS (1.5), along with the
corresponding complex coefficients (1.6), and see how (1.5)-(1.6) behave if we take
the limit T → ∞. To this end, let’s first focus on the coefficients cn : as is clear from
their expression (1.6), they depend on the index n only through the exponential
term e−2niπt/T inside the integral. In particular, these coefficients hence only depend
on n through the quantity 2nπ/T : let’s label kn this quantity, i.e.

2nπ
kn ≡ . (1.7)
T

We can thus alternatively write the coefficient cn as

 T /2
1
cn = dt e−ikn t f (t) , (1.8)
T −T /2

and we can in turn relabel it as

cn = e
ck n , (1.9)

where we introduced the coefficients

 T /2
1
ck n =
e dt e−ikn t f (t) . (1.10)
T −T /2

The only difference between cn and e ckn is that the coefficients cn are labeled by an
integer n, while the coefficients e
ckn are labeled by real numbers kn = 2nπ/T . The
advantage of these new coefficients eckn is that they are more suited to study the limit
T → ∞. More precisely, let’s consider two successive values of the integer index n,
say n and n + 1. Let’s now look at the difference kn+1 − kn , which we call ∆k, we
have

2(n + 1)π 2nπ 2π

∆k ≡ kn+1 − kn = − = , (1.11)
T T T
CHAPTER 1. GENERAL IDEA: FROM FOURIER SERIES. . . 4

and thus, in the limit T → ∞,

2π
lim ∆k = lim = 0. (1.12)
T →∞ T →∞ T

Therefore, in the limit T → ∞, the discrete real quantities kn can be viewed as

becoming an essentially continuous quantity k:
 T /2
T →∞ 1
ck n
e −→ e
ck = dt e−ikt f (t) , (1.13)
T −T /2

which we emphasize is not a true mathematical limit since we still have T in the
right-hand side.
Now, note that if we brutally take the limit T → ∞ in the right-hand side
of (1.13), we have the annoying factor 1/T that, in the limit T → ∞, goes to
zero. So to get rid of this difficulty, well let’s just somehow artificially remove it. . .
simply by multiplying by T so as to cancel the 1/T factor! By doing this, we hence
construct a function, say fe, of a continuous variable k, and we have

fe(k) ≡ lim T e
ck , (1.14)
T →∞

that is with (1.13)

 T /2  T /2
1 −ikt
fe(k) = lim T dt e f (t) = lim dt e−ikt f (t) ,
T →∞ T −T /2 T →∞ −T /2

that is simply
 ∞
fe(k) = dt e−ikt f (t) , (1.15)
−∞

which we shall call the Fourier transform of the function f .

This was for the behavior of the Fourier coefficients (1.6) in the limit T → ∞,
but what happens for the FS (1.5) itself in this limit? Well, we just write (again,
with no claim of mathematical rigor) in view of (1.7) and (1.9)
X X
f (t) = cn e2niπt/T = ckn eikn t ,
e
n kn

that is with (1.11)

X ∆k X T
f (t) = ckn eikn t
e = ckn eikn t ∆k ,
e
kn
∆k k
2π
n
CHAPTER 1. GENERAL IDEA: FROM FOURIER SERIES. . . 5

and thus in the limit T → ∞, where kn → k,

X T 1 X
f (t) = lim ckn eikn t ∆k =
e lim ck eikt ∆k ,
Te
T →∞
k
2π 2π T →∞
k
n

that is with (1.14)

1 X
f (t) = lim fe(k)eikt ∆k ,
2π T →∞ k

and thus, since ∆k → 0 in the limit T → ∞, we can write the sum as an integral,
to get
 ∞
1
f (t) = dk fe(k)eikt , (1.16)
2π −∞

which we shall call the inverse Fourier transform of fe(k).

We again emphasize that what we did here to obtain (1.15) and (1.16) is of
course very heuristic and highly doubtful from the point of view of mathematical
rigor: we took limits without real care, went from discrete indices to continuous vari-
ables, etc. . . Here the main aim was to have a grasp at how the expressions (1.15)
and (1.16) emerge when we start from the FS (1.5) and see, roughly, what this
FS should give in the limit T → ∞. Now we’ll just take the expressions (1.15)
and (1.16) as our starting point for discussing this important tool of mathematical
analysis that is the FT.
Chapter 2

Mathematical definition of the

Fourier transform

Here we give the mathematical definition of the FT. Let’s say from the very begin-
ning that there exist several slightly different conventions regarding how we math-
ematically express the FT of a function f : roughly speaking, these various conven-
tions differ in how they incorporate a factor of 2π. Therefore, to be complete and
in order to gain familiarity with the different conventions that may be used in vari-
ous courses, textbooks or fields of science and technology, here we’ll explicitly write
down what are arguably the three most frequently used conventions in section 2.1.
Since it would be pointless to separately study the properties of each of these slightly
different definitions of the FT, we’ll emphasize in section 2.2 which convention we
shall follow in this course.
REMARK ON NOTATION: before we give the definition of a FT, let’s also say
that there are many different notations that are commonly used to denote the FT
of a function f (t), such as

F [f (t)] (k) , F(f ) , F , fˆ(k) , fe(k) , (2.1)

and others may of course be found. We will use these notations interchangeably,
depending on which one is the most convenient depending on the context.

2.1 Frequent conventions for the Fourier transform

We consider a real- or complex-valued function f (t) of a real variable t. Let’s start
with the convention that actually arises directly from investigating the limit of the
FS as T → ∞, i.e. what we did above in chapter 1 [see equations (1.15) and (1.16)].

6
CHAPTER 2. MATHEMATICAL DEFINITION OF. . . 7

CONVENTION 1: the FT F [f ] (k) of f is given by

 ∞
F [f ] (k) = dt e−ikt f (t) , (2.2)
−∞

while the inverse FT of F is given by

 ∞
−1 1
f (t) = F [F [f ]] (t) = dk eikt F [f ] (k) . (2.3)
2π −∞

The equations (2.2)-(2.3) have a slightly annoying feature: the inverse FT (2.3) has
a factor 1/2π, whereas the FT (2.2) doesn’t. In other words, the FT (2.2) and the
inverse FT (2.3) are not symmetric: the advantage of the two other conventions that
we discuss below is that the FT and its inverse are symmetric.

CONVENTION 2: the FT F [f ] (k) of f is given by

 ∞
F [f ] (k) = dt e−2iπkt f (t) , (2.4)
−∞

while the inverse FT of F is given by

 ∞
−1
f (t) = F [F [f ]] (t) = dk e2iπkt F [f ] (k) . (2.5)
−∞

Note indeed that we no longer have any factor 1/2π in (2.5) as compared to the
inverse FT (2.3) in the first convention. As compared to (2.2)-(2.3), the equa-
tions (2.4)-(2.5) are now symmetric, but now involve the terms ±2iπkt in the expo-
nentials rather than merely ±ikt.

CONVENTION 3: the FT F [f ] (k) of f is given by

 ∞
1
F [f ] (k) = √ dt e−ikt f (t) , (2.6)
2π −∞

while the inverse FT of F is given by

 ∞
−1 1
f (t) = F [F [f ]] (t) = √ dk eikt F [f ] (k) . (2.7)
2π −∞

Here again the equations (2.6)-(2.7) are symmetric. While they involve the terms
±ikt (similarly to convention 1 above) in the exponentials, now both the FT (2.6)
√
and the inverse FT (2.7) have a factor 1/ 2π.
CHAPTER 2. MATHEMATICAL DEFINITION OF. . . 8

Of course, none of the above conventions is fundamentally better than the other:
to use one rather than another is mostly due to a mere personal preference and/or
some kind of tradition associated with a particular field. Furthermore, there exist
other possible conventions.

2.2 The convention followed in this course

In this course we’ll follow convention 3 above, which has the advantage, as compared
to convention 1, of exhibiting a certain symmetry between the FT and its inverse.
Therefore, from now on we say that:

The FT F [f ] (k) of the function f (t) is defined by

 ∞
1
F [f ] (k) ≡ √ dt e−ikt f (t) , (2.8)
2π −∞

while the inverse FT is given by

 ∞
−1 1
f (t) = F [F [f ]] (t) = √ dk eikt F [f ] (k) . (2.9)
2π −∞

Since the FT of a function f (t) is, by definition, an integral over R of f times

the complex exponential e−ikt , it may well be the case that this integral actually
does not exist. In this case, the FT itself hence does not exist. While there exist
mathematical theorems that ensure the existence of the FT given that f satisfies
certain conditions, to discuss this goes beyond the scope of this course.
Chapter 3

Some properties of the Fourier

transform

Here we discuss (and prove) a certain number of useful properties of the FT.

3.1 Linearity of the Fourier transform

Given two functions f (t) and g(t) (a priori complex-valued) and any two complex
numbers α and β, we have

F [αf + βg] (k) = αF [f ] (k) + βF [g] (k) . (3.1)

Proof: Because integration is a linear operation, we get from the definition (2.8)
of the FT
 ∞
1
F [αf + βg] (k) = √ dt e−ikt [αf (t) + βg(t)]
2π −∞

 ∞  ∞
1 −ikt 1
= α√ dt e f (t) + β √ dt e−ikt g(t) ,
2π −∞ 2π −∞

where we recognize in the right-hand sides the FT F [f ] and F [g], that is

F [αf + βg] (k) = αF [f ] (k) + βF [g] (k) ,

which is indeed precisely (3.1). ■

9
CHAPTER 3. SOME PROPERTIES OF THE. . . 10

3.2 Fourier transform of the complex conjugate

Let f be a complex-valued function, and f ∗ denote its complex conjugate. Then we
have

F [f ∗ (t)] (k) = {F [f (t)] (−k)}∗ . (3.2)

Proof: From the definition (2.8) of the FT we have

 ∞
∗ 1
F [f (t)] (k) = √ dt e−ikt f ∗ (t)
2π −∞

 ∞
1 ∗
dt eikt f (t)

=√
2π −∞

  ∞ ∗
1 ikt
= √ dt e f (t)
2π −∞

  ∞ ∗
1 −i(−k)t
= √ dt e f (t)
2π −∞

= {F [f (t)] (−k)}∗ ,

which is indeed precisely (3.2). ■

3.3 Fourier transform of f (−t)

Let f (−t) be the symmetric of f (t) with respect to t = 0, we have

F [f (−t)] (k) = F [f (t)] (−k) . (3.3)

Proof: From the definition (2.8) of the FT we have

 ∞
1
F [f (−t)] (k) = √ dt e−ikt f (−t) .
2π −∞

With the change of variable

t → t′ = −t ,

so that

dt = −dt′ and t = −t′ ,

CHAPTER 3. SOME PROPERTIES OF THE. . . 11

we get
 −∞
1 ′
F [f (−t)] (k) = √ (−dt′ ) e−ik(−t ) f (t′ )
2π ∞

 ∞
1 ′
=√ dt′ eikt f (t′ )
2π −∞

 ∞
1
=√ dt e−i(−k)t f (t) = F [f (t)] (−k) ,
2π −∞

which is indeed precisely (3.3). ■

3.4 Fourier transform of f (t − a)

Let a ∈ R be a constant, then f (t − a) represents the function f (t) that has been
translated by the distance a, and we have

F [f (t − a)] (k) = e−ika F [f (t)] (k) . (3.4)

Proof: From the definition (2.8) of the FT we have

 ∞
1
F [f (t − a)] (k) = √ dt e−ikt f (t − a) .
2π −∞

With the change of variable

t → t′ = t − a ,

so that

dt = dt′ and t = t′ + a ,

we get
 ∞
1 ′
F [f (t − a)] (k) = √ dt′ e−ik(t +a) f (t′ )
2π −∞

 ∞
−ika 1
=e √ dt e−ikt f (t)
2π −∞

= e−ika F [f (t)] (k) ,

which is indeed precisely (3.4). ■

CHAPTER 3. SOME PROPERTIES OF THE. . . 12

3.5 Fourier transform of f (at)

Let a ∈ R be a constant, a ̸= 0. Then we have
 
1 k
F [f (at)] (k) = F [f (t)] . (3.5)
|a| a

Proof: From the definition (2.8) of the FT we have

 ∞
1
F [f (at)] (k) = √ dt e−ikt f (at) . (3.6)
2π −∞

In view of changing the integration variable from t to at, we distinguish the two
cases a > 0 and a < 0.

Case a > 0: in this case a = |a|, and we make the change of variable

t → t′ = at = |a|t ,

so that

dt′ t′ t′
dt = and t= = ,
|a| |a| a

and thus we get for (3.6)

 ∞
1 dt′ −ik t′
F [f (at)] (k) = √ e a f (t′ )
2π −∞ |a|
 ∞
1 1 k
= √ dt e−i a t f (t)
|a| 2π −∞

 
1 k
= F [f (t)] ,
|a| a

that is indeed (3.5) for a > 0.

Case a < 0: in this case a = −|a|, and we make the change of variable

t → t′ = at = −|a|t ,

so that

dt′ t′ t′ t′
dt = − and t=− = = ,
|a| |a| −|a| a
CHAPTER 3. SOME PROPERTIES OF THE. . . 13

and thus we get for (3.6)

 −∞
dt′
 
1 t′
F [f (at)] (k) = √ − e−ik a f (t′ )
2π ∞ |a|
 ∞
1 1 k
= √ dt e−i a t f (t)
|a| 2π −∞

 
1 k
= F [f (t)] ,
|a| a

that is indeed (3.5) for a < 0.

Since we showed that (3.5) is true for both a > 0 and a < 0, we hence showed
that (3.5) is indeed true for any nonzero a ∈ R. ■