Proofs of Parsevals Theorem & the Convolution Theorem

(using the integral representation of the -function)

The generalization of Parsevals theorem

The result is

1
f ()g() d
2

This has many names but is often called Plancherels formula.

f (t)g(t) dt =

(1)

The key step in the proof of this is the use of the integral representation of the -function
( ) =

1
2

ei d

or

() =

1
2

We firstly invoke the inverse Fourier transform

f (t) =

1
2

ei d .

(2)

f ()eit d

(3)

and then use this to re-write the LHS of (1) as

Z

Z 
Z
1

f (t)g(t) dt =

it

f ()e



1
2

f (t)f (t) dt =

1
2

2 Z Z

f ()g( )

Z
|

g( ) e

Re-arranging the order of integration we obtain

Z

0 i 0 t

i( 0 )t

dt

{z

U se deltaf n here

dt .

d 0 d .

(4)

(5)

The version of the integral representation of the -function we use in (2) above is
( 0 ) =

1
2

eit( ) dt .

(6)

Using this in (5), we obtain

Z

f (t)g(t) dt =
=

Z

1
f ()
g( 0 ) ( 0 ) d 0 d
2

Z
1
f ()g() d .
2

(7) comes about because of the general -function property

F (

0 )(

(7)

0 ) d 0 = F ().

Parsevals theorem (also known as the energy theorem)

Taking g = f in (1) we immediately obtain

Z

1
|f (t)| dt =
2

|f ()|2 d .

(8)

The LHS side is energy in temporal space while the RHS is energy in spectral space.
R
dt
Example: Sheet 6 Q6 asks you to use Parsevals Theorem to prove that
= /2.
(1+t2 )2
The integral can be evaluated
by
the
Residue
Theorem
but
to
use
Parsevals
Theorem
you
will
R eit dt
H eiz dz
need to evaluate f () = 1+t2 . To find this, construct the complex integral C 1+z 2 and
take the semi-circle C in the upper (lower) half-plane when < 0 (> 0). The answers are e
when < 0 and e when > 0. Then evaluate the RHS of (8) in its two parts.
1

The Convolution theorem and the auto-correlation function

The statement of the Convolution theorem is this: for two functions f (t) and g(t) with Fourier
transforms F [f (t)] = f () and F [g(t)] = g(), with convolution integral defined by1
f ?g =

f (u)g(t u) du ,

(10)

then the Fourier transform of this convolution is given by

F (f ? g) = f () g() .

(11)

To prove (11) we write it as

F (f ? g) =

it

Z

f (u)g(t u) du

dt .

(12)

Now define = t u and divide the order of integration to find

F (f ? g) =

eiu f (u) du

ei g( ) d = f () g() .

(13)

This step is allowable because the region of integration in the u plane is infinite. As we shall
later, with Laplace transforms this is not the case and requires more care.
The normalised auto-correlation function is related to this and is given by
(t) =

f (u)f (t u) du
R
2
|f (u)| du

(14)

It makes no difference which way round the f and the g inside the integral are placed: thus we could write
f ?g =

f (t u)g(u) du .

(9)