Fourier Transforms

So far we have only found Fourier series

expansions for periodic functions. Suppose we
have a function that is not periodic and wish to
decompose it into a superposition of sine waves.
How do we do it?
We will try to gain some insight into this
problem by considering an example.
Let’s reconsider a square wave that has positive
and negative regions of different widths.

We will find the Fourier expansion for this

function and then allow L to become large
while w is held constant. In the limit that
L → ∞, f ( x ) will consist of a single “top hat”,
and will no longer be a periodic function.

1
The Fourier expansion in the interval (− L, L) is
a0 ∞  nπ x nπ x 
f ( x ) = + ∑ an cos + bn sin
2 n=1  L L 
From the graph we see that f ( x ) is an even
function. So there can be no sine waves in the
expansion, and,
bn = 0.
Next we calculate
1 L nπ x
an = ∫ f ( x ) cos dx for n = 0,1,2...,
L −L L

which gives
4w 4  nπ w 
a0 = − 2; an = sin , n = 1,2...
L nπ  L 

2
Let’s consider the case that L = 2w then

a0 = 0

a2m = 0 ,

4(−1)m−1
a2m−1 = , m = 1,2...
(2m −1)π

4 4 4 4
i.e. an = 0, ,0,− ,0, ,0, − ,.. n = 0,1,..
π 3π 5π 7π
Let’s introduce a new variable
nπ nπ
k= = .
L 2w
Then we have
∞
f ( x) = ∑ an cos kn x .
n=1
Now lets plot an as a function of kn .

3
Next let us instead make L = 4w. Then

a0 = −1,

4  nπ 
an = sin , n = 1,2...
nπ  4 

nπ nπ
Plotting these coefficients against k = = ,
L 4w

4
And now if we set L = 8 w then
3
a0 = − ,
2
4  nπ 
an = sin  
nπ 8
Ploting the Fourier coefficients against
nπ nπ
k= = we obtain
L 8w

5
If we were to continue to increase the ratio of L to
w then we would find that:
• The crossing point of the “envelope” curve does
not move on the k axis
• The magnitude of an decreases
• The density of points on the k axis continues to
increase and tends towards a continuum of values
As we let L tend to ∞ we are making a Fourier
expansion of a single “top hat” which will be valid
for all values of x .
We can replace the summation by an integral
a0 ∞ nπ x a0 ∞
+ ∑ an cos → + ∫ A(k ) cos kx dk.
2 n=1 L 2 0
where
La(n )
A(k ) =
π
1 ∞
= ∫ f ( x ) cos kx dx
π −∞
a0
and = −1 is the mean value of the function.
2
We have made a Fourier integral expansion of the
function f ( x ) and the function A(k ) is itsFourier
Transform .

6
Fourier’s Integral Theorem
In our example there were no sine terms because our
original function was an even function. For a more
general function let’s try an expansion
∞
f ( x) = ∫ {A(k ) cos kx + B(k )sin kx}dk
0
We can do this if
1. f ( x ) satisfies the Dirichlet conditions for all
values of x .
∞
2. ∫ f (x) dx converges to a finite value. This
−∞
implies that the average value of the function
must be zero!
As for Fourier series, at a point of discontinuity the
value of the Fourier expansion is the average of the
values on either side of the discontinuity.
Can we find expressions for the Fourier
coefficients A(k ) and B(k )?
We use a similar approach as for Fourier series:
multiply both sides by cos k ′x and integrate
w.r.t. x between ±∞.

7
+∞
∫ f ( x ) cos k ′x dx
−∞
+∞ ∞
= ∫ ∫ {A(k ) cos kx + B(k ) sin kx}dk cos k ′x dx
−∞ 0

8
 1 ∞
A(k ) = ∫ f ( x ) cos kx dx
π −∞
and we made use of the result
∞
∫ cos ((k − k ′ )x )dx = 2 πδ (k − k ′ )
−∞
9
If instead we were to multiply both sides by
sin k ′x and integrate then we could obtain
1 ∞
B(k ) = ∫ f ( x )sin kx dx
π −∞
In summary, the Fourier integral expansion of
the function f ( x ) is given by
∞
f ( x) = ∫ {A(k ) cos kx + B(k )sin kx}dk
0

where
1 ∞
A(k ) = ∫ f ( x ) cos kx dx
π −∞
1 ∞
B(k ) = ∫ f ( x )sin kx dx
π −∞
These latter expressions are in fact Fourier
transforms of the function f ( x ).
This form for the Fourier integral expansion
illustrates the similarity to the Fourier series
expansion. Notice how the Dirac delta function
appeared in analogy to the Kronecker delta.
However this is not the form that is most
commonly used. Rather we need to rewrite it in
terms of exponential functions.
10
Exponential form of Fourier’s Integral
Theorem
In the equations that we have just derived we
may replace the sine and cosine functions with
exponentials using the Euler formulae.

11
We then obtain
∞ −ikx
f ( x) = ∫ F (k )e dk
−∞
where
1 ∞
F(k ) = ∫ f ( x )e ikx
dx
2 π −∞

1
Actually we are free to share the factor of
2π
between f ( x ) and F(k ) as we like. A
symmetric choice is probably the most common
and from now on we will us the following
definitions:
F(k ) as The Fourier Transform of f ( x ) where
1 ∞ −ikx
f ( x) = ∫ F (k )e dk
2 π −∞
and f ( x ) is the inverse Fourier Transform of
F(k ), where
1 ∞
F(k ) = ∫ f ( x )e dx
ikx
2 π −∞

12
Example
Find the Fourier transform of
 1 , X −w< x < X+w
f ( x ) = 2w .
 0, x < X − w, x > X + w

13
Example
Find the Fourier transform of
1− x 2 , x ≤1
f ( x) =  .
 0, x >1

14
 The Dirac Delta Function
We found that the Fourier transform of a ‘top
hat’ with area equal to unity is
1
F(k ) = sinc(kw)eikX .
2π
If we squeeze the ‘top hat’, then as w → 0 , f ( x )
tends to the Dirac delta function δ ( x − X ).
If we let w → 0 in the expression for F(k ) , we
must obtain the Fourier Transform of δ ( x − X ).
So, as w → 0, sinc(kw) → 1, for all values of k ,
and so
1 ikX
F(k ) → e
2π
In summary
1 ∞ 1 ikX
F(k ) = ∫ δ ( x − X )e dx =
ikx
e .
2 π −∞ 2π
The Fourier transform of the Dirac delta
function is an exponential.
Consider next the inverse transform
1 ∞ −ikx
f ( x) = ∫ F (k )e dk
2 π −∞
15
which gives the useful result
1 ∞ ik ( X− x )
δ (x − X ) = ∫e dk
2 π −∞
We can show also that the Dirac delta function is the
Fourier transform of an exponential function.
1 iKx
If f ( x ) = e then
2π
1 ∞ 1 iKx ikx
F(k ) = ∫ e e dx
2 π −∞ 2 π
1 ∞ i(k + K )x
= ∫e dx
2 π −∞
= δ (K + k)
This should not surprise us: the exponential function
represents a single plane wave.
The delta function allows us to pick out this single
plane wave in the Fourier expansion as follows:
1 +∞ −ikx
f (x) = ∫ F (k )e dk
2 π −∞
1 +∞ −ikx
= ∫ δ (K + k )e dk .
2 π −∞
1 iKx
= e
2π

16
Example
Consider the Gaussian function
 x2 
f ( x ) = exp − 2  ,
 a 
that has a half width at half maximum of
∆x = a ln 2 .
The Fourier transform is given by

17
 π  a2k 2 
F(k ) = a exp − 
2  4 
for which
2
∆k = ln2 .
a
Hence we find that
∆x ∆k = 2 ln2,
which is independent of the constant a that
determines the width of both f ( x ) and F(k ).
It is a general feature of Fourier transforms that
if we make the function f ( x ) wider, then F(k )
becomes narrower.
The Heisenberg Uncertainty Principle provides
a physical example of this phenomenon.
In fact the Gaussian is rather special: it is the
function that minimises the product ∆x∆k.

18
The Convolution Integral
The convolution of two functions f ( x ) and
g( x ) is defined as
1 ∞
f ∗g = ∫ f (u)g( x − u )du
2 π −∞

Example: Consider the functions

1, x ≤ 1
f ( x) = 
0, x > 1
1 − x, 0 ≤ x ≤ 1
and g( x ) = 
 0, x < 0, x > 1

19
The convolution is given by
0, x >2
 1 ( x − 2)2 , 1< x ≤ 2
 2 2 π
1
f ∗g =  , 0 < x ≤1
2 2π
 1
2 2π ( 1− x 2
)
, −1< x ≤ 0
0, x ≤ −1
Does f ∗g = g∗ f ?

20
The Convolution Theorem
The Fourier transform of the convolution of
f ( x ) and g( x ) is equal to the product of the
Fourier transform of f ( x ) and the Fourier
transform of g( x ).
F { f ∗g} = F { f }F {g}
where F { f } denotes the Fourier transform of
f ( x ).

21