0.

1 The Inversion Formula

The characteristic function corresponds to a family of distribution obtained by
adding an arbitrary constant to a distribution function of a random variable.
The inversion formula is a tool that can be used to get back the original
distribution function on the entire real line if the characteristic function is
known.
Theorem
Let F(x) and φx (t) be the cumulative distribution function and the character-
istic function of the random variable X respectively, then for given real numbers
the inversion formula is defined as
ZC
1 e−ita − e−itb
F(b) − F(a) = `im φx (t)dt
C→∞ 2π it
−C

Proof
1
RC e−ita −e−itb
Let Ic = F(b) − F(a) = 2π it
φx (t)dt
−C
(t) e−ita −e−itb
First we need to show that ϕx ≤ 1 and that it
is bounded.

E eitx ≤ E eitx


|ϕx (t)| =
≤ E |Costx + i Sint x|
≤ E (Cost x + i Sint x) (Cost x + i Sint x)
≤ E Cos2 t x + i Sin2 t x

i.e. |ϕx (t)| ≤ 1

0.1. THE INVERSION FORMULA

Zb
e−ita − eitb
Ic = ≤ e−itx dx
it
a
Zb
≤ dx e−itx
a
Zb
≤ dx Since e−itx ≤ 1
a
≤ b−a hence bounded.

Now it is possible to apply the Fubini’s theorem to Ic as

1
RC e−ita −e−itb
R∞
Ic = 2π it
eitx dF(x) dt; (a < b)
−C −∞
RC ∞
1 e−ita −e−itb
R itx
= 2π it
e dF(x) dt
−C −∞

1
RC eit(x−a) −eit(x−b)
= 2π it
dt) dF(x)
−C

Where eit(x−a) − eit(x−b) can be written as

Cost (x − c) + i Sin t (x − a) − Cost (x − b) + i Sin t (x − b)

1
R∞ RC Cost (x−a)+i Sin t(x−a)−Cost (x−b)+i Sin t(x−a) dt)
Ic = 2π it
dF(x)
−∞ −C

multiply numerator and denominator by i

Z∞  
1 2i Cost (x − a) + i Sin t (x − a) − Cost (x − b) + i Sin t (x − a) dt)
= dF(x)
2π i it
0

as c −→ ∞
Z∞ Z∞
1
Ic = ... i Cost (x − a) − Sin t (x − a) + i Cost (x − b) − Sin t (x − b) dt]dF(x)
π
−∞ 0

ii
0.1. THE INVERSION FORMULA

Since the integration of a cosine function gives a sine function that will later
vanish
Z∞ Z∞  
1 Sin t (x − a) − Sin t (x − b)
Ic = dt dF(x)
π t
−∞ 0

R∞ Sin v π
By complex integration v
dv = 2
0
R∞ Sin P v π
and v
dv = 2
Sgn P where
0





 1 if P > 1


Sgn P = 0 if P = 0





 −1 if
 P <1


− π2 + π




 2
= 0; x < a, x < b











π
= π/2;




 0+ 2
x = a, x < b











Z∞ 
 π/ + π/ = π; a<x<b
Sin t (x − a) − Sin t (x − b)  2 2
dt =
t 

0 





−π/2 + 0 = π/2;

x < a, x = b















π/ − π/ = 0;

x > a, x > b

2 2










iii
0.1. THE INVERSION FORMULA

Ra Ra Rb− Rb
Ic = 0.dF(x) + π/ dF(x) + 0.dF(x) + −π/ dF(x)
2 2
−∞ −a −∞ −a

= − π2 F(a) − F(a−1) + π
   
2
F(b) − F(a−1)

π
 
= 2
F(b) − F(a)

1

= F(a+0) − F(a−0) + F(b+0) − F(b−0) + 2
F(a+0) − F(a−0)

= F(b) − F(a)

If a and b are points of continuity of F .

Corollary

(1) The distribution function F (x) of a random variable and its characteristic
function ϕx (t) determines each other.

(2) If F (x) is the distribution function of a random variable then by defini-

tion, it determines the characteristic function ϕx (t) uniquely.

Example 1:
n
If φx (t) = (q + ρeit ) is the characteristic function of the binomial random
variable, obtain the probability density function of a random variable X.
Solution
Zπ
1
F(x) = e−itx φx (t)dt
2π
−π

iv
0.1. THE INVERSION FORMULA

n
Given φx (t) = (q + peit ) ,

1
Rπ n
F(x) = 2π
eitx (q + peit ) dt
−π  
Rπ n  n 
1
eitx eitj   pj q n−j dt
P
= 2π  
−π j=0
j
 
Rπ n  n 
1
eitx eitj   pj q n−j dt
P
= 2π  
−π j=0
j
 
n  n  Rπ
1   pj q n−j Cos t (x − j) − i Sin t (x − j)dt
P
= 2π  
j=0 −π
j

Let u = t (x − j)

du du
dt
= (x − j) ⇒ dt = (x−j)
 
n  n  Rπ
1   ρj q n−j 2 Cos t (x − j) − i Sin t (x − j)dt
P
= 2π  
j=0 0
j

But as n → ∞

π ∞
R∞ Sin t(x−j) R∞ Cos t(x−j)
= (x−j)
−i (x−j)
0 0
0 0

=π−0

v
0.1. THE INVERSION FORMULA
 
n  n 
π   ρj q n−j
P
π  
j=0
j

`im
n→∞  
∞  n 
  ρj q n−j
P
=  
j=0
j
 
 n  j n−j
f(x) = 
 ρ q

j

Example 2:
−1
Let p (1 − qeit ) is the characteristic function of the Geometric random vari-
able, obtain the probability density function of a random variable X.
Solution

−1
QX (t) = p 1 − qeit
Z π
1
−1 −itx
F (x) = P 1 − qeit e dt
2π −π
Z π
p 1
= e−itx dt
2π −π 1 − qeit
Z π X n
!
p
= ·2 qeit e−itx dt
2π 0 i=1
n Z π
p X
= qj e−it(x−j) dt
π i=1 0
n
pX j π
Z
= q [cos tt(x − j) − i sin t(x − j)] dt
π i=1 0

but
π π
sin t(x − j) cos t(x − j)
− =π
(x − j) 0 (x − j) 0

vi
0.1. THE INVERSION FORMULA

n
qj
P
p ∞
j=1 X
∴ F (x) = ·π = pq j
π j=1
j
∴ f (x) = pq (0.1.1)

vii
0.1. THE INVERSION FORMULA

Exercise
Obtain the characteristic function for the following joint density function:

(i) f (x, y) = exp {− (x + y)}

1
exp − 12 (x2 + y 2 ) ;

(ii) f (x, y) = 2π −∞ < x, y < ∞

(iii) P (x, y) =

x
y
0 1 3

1 0.1 0.2 0.3

2 0.1 0.05 0.25

(iv) Obtain m3 and m4 what are your observation (m2 , m3 , m5 ) able to equal
to zero

(v). Obtain the mean and variance for each of the following characteristic
functions:
(t)
i ϕx = exp i µ t − 21 τ 2 t2


ii ϕx = α (α − it)−1
(t)

= α1 , α2 (α1 − it1 )−1 (α2 − it2 )−1

(t , t2 )
iii ϕxy1

(vi). Show with appropriate theorem that the variance of a one-point distri-
bution is zero.

(vii). Define the one-point distribution and obtain the cumulative distribution
function.

viii