1 The Fourier Transform and Its Inverse.

The Fourier transform of a function f 2 L1 (R) is de…ned to be

Z 1
fb( ) = f (x) e i x dx;
1

for each 2 R. The integral de…ning fb( ) is convergent (and therefore mean-
ingful) because
Z 1 Z 1 Z 1
f (x) e i x dx f (x) e i x dx = jf (x)j dx < 1:
1 1 1

The inverse Fourier transform of a function g = g ( ) is de…ned to be

Z 1
1
G( ) = g ( ) ei x d :
2 1

Theorem 1 Let the function f 2 L1 (R) be piecewise smooth on R, and let

fb( ) be its Fourier transform. Then the inverse Fourier transform of fb( ) is
equal to the function de…ned by
1
f x+ + f x
2
for every x 2 R. Thus
Z 1
1 1
fb( ) ei x d = f x+ + f x ;
2 1 2

for every x 2 R, and at points x of continuity of f ,

Z 1
1
fb( ) ei x d = f (x) :
2 1

To establish this result, we need to prove the following two results

Theorem 2 (Riemann-Lebesgue lemma) Let the function f 2 L1 (R) be piece-

wise smooth on R, and let fb( ) be its Fourier transform. Then fb( ) ! 0 as
j j ! 1:
Proof. Since e i x = cos ( x) i sin ( x) ; it su¢ ces to prove that
Z 1 Z 1
f (x) cos ( x) dx ! 0; and f (x) sin ( x) dx ! 0;
1 1

as ! 1: We prove only the …rst result; the second is proved in a very similar
way to the …rst. Choose " > 0: Because f 2 L1 (R), there exists an M > 0 such
that Z Z M Z 1
1
jf (x)j dx = jf (x)j dx + jf (x)j dx < ":
jxj>M 1 M 2

1
Thus
Z Z Z
1
f (x) cos ( x) dx jf (x) cos ( x)j dx ": jf (x)j dx <
jxj>M jxj>M jxj>M 2
(1)
Since the function is piecewise continuous on the interval M x M , it
is a bounded function so that there is a number B > 0 so that f (x) B for
M x M . Moreover, it is Riemann integrable and there exists a partition
of m points of the interval, of form,
M = x0 < x1 < : : : < xm 1 < xm = M
such that the step function (piecewise constant function) h = h (x) de…ned by
8
>
> f (x1 ) ; x0 < x x1 ;
<
f (x2 ) ; x1 < x x2 ;
h (x) =
>
> : : : :::
:
f (xm ) ; xm 1 < x xm ;
satis…es
Z M
1
jf (x) h (x)j dx < ":
M 2
(In other words we may approximate f by the step function h - this is a conse-
quence of the de…nition of the Riemann integral.) Now
Z M Z M Z M
f (x) cos ( x) dx = [f (x) h (x)] cos ( x) dx + h (x) cos ( x) dx:
M M M

The …rst term on the right hand side satis…es

Z M Z M Z M
1
[f (x) h (x)] cos ( x) dx j[f (x) h (x)] cos ( x)j dx jf (x) h (x)j dx < ":
M M M 2
(2)
The second term on the right hand side can be calculated, when > 0;
Z M Z x1 Z x2 Z xm
h (x) cos ( x) dx = h (x) cos ( x) dx + h (x) cos ( x) dx + : : : + h (x) cos ( x) dx
M x0 x1 xm 1
Z x1 Z x2 Z xm
= f (x1 ) cos ( x) dx + f (x2 ) cos ( x) dx + : : : + f (xm ) cos ( x) dx
x0 x1 xm 1

sin ( x1 ) sin ( x0 ) sin ( xm ) sin ( xm 1)

= f (x1 ) + : : : + f (xm )
m
X sin ( xk ) sin ( xk 1)
= f (xk ) ;
k=1

and so
Z M m
X sin ( xk ) sin ( xk 1) 2mB
h (x) cos ( x) dx jf (xk )j : (3)
M k=1

2
Putting equations (1), (2) and (3) together gives
Z 1
1 1 2mB 2mB
f (x) cos ( x) dx "+ "+ ="+ : (4)
1 2 2
Equation (4) holds for each …xed " > 0 and > 0; and the number m depends
only on ". Thus we deduce
Z 1
f (x) cos ( x) dx "; as ! 1: (5)
1

But equation (5) holds for every " > 0 so we deduce

Z 1
lim f (x) cos ( x) dx = 0:
!1 1

This completes the proof.

Before stating the next result, we introduce the function Da , where a is a
real number, by
sin (ax)
Da (x) = :
x
Theorem 3 Let the function f 2 L1 (R) be piecewise smooth on R, and let
fb( ) be its Fourier transform. Then for every a > 0, the truncated inverse
Fourier transform Z a
1
fa (x) = fb( ) ei x d
2 a
is equal to the convolution (Da f ) (x) :
Proof.
Z a
1
fa (x) = fb( ) ei x d
2 a
Z a Z 1
1
= f (y) e i y dy ei x d
2 a 1
Z 1 Z a
1
= ei (x y) d f (y) dy
2 1 a
Z 1
1 eia(x y) e ia(x y)
= f (y) dy
2 1 i (x y)
Z
1 1 sin a (x y)
= f (y) dy = (Da f ) (x) :
1 (x y)

The graph of Da (x) is shown for various values of a; notice that as a in-
creases, the function becomes more and more "concentrated" near the origin.
Thus providing that f is continuous near x, we might expect that
Z 1
(Da f ) (x) = Da (y) f (x y) dy
1

3
to be approximately equal to
Z 1
f (x) Da (y) dy;
1

at least as a ! 1: It is a fact that

Theorem 4
Z 1 1; if a > 0;
Da (y) dy = 0; if a = 0;
1 1; if a < 0:

This explains why the inverse Fourier transform of fb( ) ; which is really the
limit of its truncated inverse Fourier transform fa (x) as a ! 1; should equal
the original function f (x), at least at points of continuity.We can now prove the

y=sin(ax)/x, various values of a

4
a=1
3.5 a=2
a=4
3

2.5

1.5
y

0.5

-0.5

-1
-15 -10 -5 0 5 10 15
x

main result.

4
Proof. We have
Z a Z 1
1
fa (x) = fb( ) ei x d = (Da f ) (x) = Da (y) f (x y) dy
2 a 1
Z 0 Z 1
= Da (y) f (x y) dy + Da (y) f (x y) dy
1 0
Z 1 Z 0
= Da (t) f (x + t) dt + Da (t) f (x + t) dt:
0 1

We claim that Z 1
1
lim Da (t) f (x + t) dt = f x+ (6)
a!1 0 2
and Z 0
1
lim Da (t) f (x + t) dt = f x : (7)
a!1 1 2
We prove only equation (6); equation (7) is proved similarly. Separate the
integral occurring in (6) into two parts:
Z 1 Z Z 1
Da (t) f (x + t) dt = Da (t) f (x + t) dt + Da (t) f (x + t) dt
0 0
Z Z
1 sin(at) 1 1 sin(at)
= f (x + t) dt + f (x + t) dt:
0 t t
De…ne the function g by

f (x + t) =t; t > ;
g (t) =
0 t :

Then g is a piecewise continuous function, and by the Riemann-Lebesgue lemma,

Z 1
g (t) sin (at) dt ! 0; as a ! 1;
1

that is, Z 1
1 sin(at)
lim f (x + t) dt = 0:
a!1 t
We must now deal with the limit
Z
1 sin(at)
lim f (x + t) dt:
a!1 0 t
The function
f (x + t) f (x+ )
h (t) =
t
is piecewise continuous on [0; ] ; and by the Riemann-Lebesgue lemma,
Z
h (t) sin (at) dt ! 0; as a ! 1:
0

5
Then
Z
1 sin(at)
lim f (x + t) dt
a!1 0 t
Z Z
1 f (x + t) f (x+ ) 1 sin(at)
= lim sin(at)dt + f x+ dt
a!1 0 t 0 t
Z Z
1 1 sin(at)
= lim h (t) sin(at)dt + f x+ dt
a!1 0 0 t
1
= f x+ ;
2
because by Theorem 4,
Z Z Z
1 sin(at) 1 a sin(u) 1 1 sin(u) 1
lim dt = lim du = du = :
a!1 0 t a!1 0 u 0 u 2

We have now established the main result on the Fourier transform and its inverse
stated in Theorem 1.