The Convolution of Gamma Distribution with A Stabilizer Constant

Applied Mathematical Sciencess, Vol. x, 20xx, no, xx, xxx-

xxxx-----------

The Convolution of Gamma Distributions with A Stabilizer Constant

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar

Faculty of Mathematical and Natural science, Andalas University

Limau Manis Padang, West Sumatra, Indonesia,

rika_okdamarlinazen@yahoo.com

Abstract. A Gamma distribution is one of continous distributions which have two parameters
namely a shape parameter of ( ) and a scale parameter of (  ) . If   1 , then the gamma
distribution will become the exponential distribution. In this paper, the new distribution will be
determined by adding a stabilizer constant on the Gamma distribution. This adding is aimed to
preserve the characteristics of its probability density function. Next, the convolution and the
characteristics of the Gamma distribution with the stabilizer constant will be found such as the
mean, variance, and moment generating function of a random variable W.

Keywords: Random variable, Gamma function, Beta function, Gamma distribution, convolution,
stabilizer constant.

1. Introduction
The probability distribution is divided into two types based on its random variable, namely
the discrete and continuous probability distributions. One of continuous probability distributions
is a gamma distribution. According to Bain [2], the name of gamma () is derived from the

gamma function denoted by  (k ) and defined by (k )   x k 1e  x dx for all k  0
0

Gamma distribution can be used to solve problems in the fields of statistics, engineering and
science. Gamma distribution plays an important role in queuing theory and reliability theory
(reliability), for example to overcome the problem of data loss. According to Rosita [10],
estimation of gamma distribution parameters can be used to deal with the loss of credit sales data
on an official site.
In the journal, Devianto et al. [3] discusses the exponential distribution and convolution of
exponential distribution with a stabilizer constant. Thus, the following gamma distribution can
also be given the stabilizer constant. The provision of the stabilizer constant is intended to
maintain the characteristics of its probability density function.

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

2. Terminologies and Notation of Gamma Distribution


The gamma function is denoted by  (k ) and is defined as (k )   x k 1e  x dx
0

for all k  0 . Suppose X 1 , X 2 ,..., X n are random variables of gamma distributions with
parameters  i , i  1,2,...., n and  i , i  1,2,...., n denoted X ~ Gamma( ,  ) has the following
probability density function
    1  x
 x e , if 0  x  ;   0,   0
f X ( x;  '  )   ( ) (2.1)

 0 , for x otherwise
Gamma distribution has the cumulative distribution function, expectation value, variance, and
moment generating function of the random variable X , as follows.
  x  1  x
( ) 0
FX ( x; '  )  x e dx


EX  


Var ( x)  2


 t 
M x (t )  1  
 

3. Probability Density Function of the Gamma Distribution with the Stabilizer Constant
Let X be a random variable that has Gamma distribution with the parameters  and  ,
and the probability density function of the random variable X shown in equation (2.1).
Moreover, a random variable W is said to have a gamma distribution with a stabilizer
constant if W ~ Gamma stabilizer( ,  ) where   0 ,   0 and 0  w  1. Suppose that a
gamma distribution is defined by changing its boundary from 0,  to be 0,1 then a new
probability density function can be formed by multiplying a stabilizer constant θ [11] that is:

   1 w
f W ( w; '  )   w e (3.1)
( )

for  ,  positive and 0  w  1, where  is a stabilizer of the gamma distribution that depends
on  and  , and is defined as:

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

 i 
 1 (3.2)
w
i  i 1  wi  i
i i e dwi
0

Equation (3.2) will be solved for the value of  i  2 so that the value θ is obtained as follows

1
 (3.3)
e  i
  i  1  1

Thus, the probability density function and the cumulative distribution function of the random
variable W are yielded as follows:

 i 2 wi e  w  i

f Wi ( wi ;  i '  ) 
e     i  1  1
i

 2, w
FW ( w; 2 '  )  , if w  0 (3.4)
e  i
  i  1  1
for 0  w  1,  i  2 and  i  0

4. Convolution of Gamma Distribution with a Stabilizer Constant for Different Parameters

Convolution of Gamma distribution with a stabilizer constants for different parameters is
given as follows.

Proposition 4.1. Suppose that there are n identically independent variables, namely Wi for
i  1,2, ... , n so that Wi has a probability density function defined as follows:
 i 2 wi e  w  i

f W ( wi ;  i '  )   
i
e   i  1  1 i

for 0  w  1,  i  2 for all i  1,2, ..., n and  i  0 . Then, the sum of random variable Wi that
n
is S n   Wi has the probability density function as follows:
i 1

 n 

 2  1 1 1  wp
f Sn s n   Psn   , ..., u1 , ... , u n1 du1 , ... , d u
   i '....., n  u1 ,...,u n 1 
 i 1    ... e  
i n 1 (4.1)
0 0

 s n  0 and f sn s n   0,  w  0, where

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

 i
2

P i 1
(4.2)
 e    
n
 n 
 i
i  1  1   2 
i 1  i 1 
and
n 1 n 1 n 1
p
 i ,....., n  u1 ,...,u n 1 
:  i  u j  i 1ui 1  u j  n 1un1  (4.3)
  j 1 i 2 j i

and
n
n 1  2

 u ji1 1  u j ,
(4.4)
1 ,...,  n u1 ,..., un 1  :
1
 n 1 ,...,  n 1  j i

for all u1 , ..., u n1  0,1

Proof. The theorem above will be proven by using mathematical induction. First, it is shown
that equation (4.1) is true for the case n  2 . Let g be a continuous function on the real
number line. The integral I 2 (g) will be calculated as
I 2 (g)  Eg w1  w2 , (4.5)
and it is defined as follows:
1 1
12  22
I 2 g     g w1  w2  (4.6)
0 0 e  
1
1  1  1 e    2
2  1  1 
By transforming the variable in equation (4.6), for example wi  yi2 , for i  1,2 , it gives the
result
12  22
 y 
 y 23 e y1 1  y2  2 dy1dy 2
1 1
I 2 g   4   g y1  y2
2 2
2 2 3
(4.7)
 e   
2 1
i
i  1  1 0 0

i 1

Next, by using spherical coordinates,

y1  r sin  n 1 ... sin  3 sin  2 sin 1 (4.8)
y 2  r sin  n 1 ... sin  3 sin  2 cos 1
y 3  r sin  n 1 ... sin  3 cos  2
...  ...
y n 1  r sin  n 1 ... sin  n  2
y n  r cos  n 1 ,

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant


where 0  r  , and 0   k  , for each k  1,... , n  1 .Then, given
2
y k 1 r
rk2  y12  ...  y k2 and rk  r , we have cos k   and sin  k   k (4.9)
rk 1 rk 1

so that the Jacobian is found by changing the variable yi with i  1,2, ... n
n 1
J  r n 1 sin nn21 sin nn32 ... sin  2  r n 1  sin k 1  k . (4.10)
k 1

Hence, equation (4.7) gives



 
  g r 
2 2 1 2
I 2 g   4 sin 1  r 2 cos 2 1 r sin 1  r cos 1 
1 2 2 3 3
(4.11)
 e   
2
i
i  1  1 0 0

i 1

e 1r sin2 1   2 r 2 cos2 1 rd dr

2

1


 
  g r 1r  sin   cos   e
1 2
 rd dr (4.12)
2 2
3  r 2 1 sin2 1   2 cos 2 1
4 1 2 2 3 3

 e    
2 1 1 1
i
i  1  1 0 0

i 1

Suppose that s 2  r 2 , and u j  sin 2  j , for all j  1,..., n  1 , so that equation (4.12) can be
written in the form of the following equation


  g s s 
  2 2 1 2
I 2 g   4 1 2 4 1

 e    
2 2 2
i
i  1  1 0 0 s2
i 1

 1  s  u   1u  2 1 
 0 e 2 1 1 2 1 u1 1  u1 2 1
du1  ds 2
 u1 1  u1 4 s 2 u1 1  u1 


  3  1  s2  1u1   2 1u1 
  
2 2 1 2
  
1
g s 2
s 0 e
1
u1 1  u1 du1  ds 2 (4.14)
B2 2,2
 
2 2 2
4 e i   i  1  1 0 0  
i 1

Based on equation (4.14), for each continuous function g which is in R , then the random
variable S 2 has the probability density function f s2 given for each positive real number s2
namely
Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

12  22 3  1  s 2  1u1   2 1u1 

1 2

f s2 s 2    
1
0 0 0
s  e
B2 2,2
u 1  u du , (4.15)
 
2 2 1 1 1
4 e   i  1  1
i 
i 1

Regarding on the properties of the probability density function, the formula above holds for
n2
The next step is to show that equation (4.1) is true for n  3 . Let g be a continuous
function on a real number line, with S 3  W1  W2  W3 or it can be written as S 3  S 2  W3 . It
will be determined
I 3 (g)  Eg s 2  w3 , (4.16)

which is defined as follows

1 1
12  22
I 3 g     g s 2  w3  s 23 (4.17)
   i  1  1
2
0 0
4 e i

i 1

12  22  32 1 1

  g s  w3 s 23 w3  e  s2 1u1   2 1u1  w33 u1 1  u1 du1  ds 2 dw3

1

 0 
 e    
3 2
 i
i  1  1 32 0 0 (4.18)
i 1

Furthermore, by transforming the variable in equation (4.18), let s 2  y12 , and w3  y 22 , .

Then it is obtained

12  22  32
  g y 
1 1
I 3 g    y 22 y16 y 22   e  y1 1u1   2 1u1  y2 3 u1 1  u1 du1  dy12 dy 22
1 2 2
2
 0 
 e   
3 1
 i
i  1  1 32 0 0

i 1 (4.19)

12  22  32
 
1 1
7 3
e 1 1 1 2 1 2 3 u1 1  u1 du1  dy1dy 2
1  y 2   u   1u  y 2 
4 0 0 
0
2 2
g y y y y

 e   
3 1 2 1 2
 i
i  1  1 32
i 1 (4.20)

In addition, the spherical coordinates in equation (4.8) can be used to solve the equation
(4.20) so that the equation (4.20) becomes

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

  
  g r r
2 2 2 1 2
I 3 g   4 1 2 3 2 11
sin 72 cos32 (4.21)
 e   
3
 i
i  1  1 32 0 0

i 1

 1 e r 2 sin22  1u1   2 1u1 r 2 cos22 3 u 1  u du  r d dr

0 1 1 1
 2

Given s3  r 2 , and u j  sin 2  j , for all j  1,..., n  1, so that equation (4.21) can be written
in the form of the following equation

12  22  32 1
I 3 g    g s s
5
(4.22)
 e   
3 3 3
 i
i  1  1 2  2  2 0
i 1

 1 1 r 2u2  1u1   2 1u1  s3 1u2 3 

u1u 2 1  u1 du1 du 2  ds3
1
  e
3

0 0 B3 2,2,2 

Based on equation (4.22) for each continuous function g , which is in R, it can be seen that
the random variable S 3 , has the probability density function f s3 , given for each positive
real number s3 , The probability density function above f s3 , can also be written as follows

 i
2
1
 3 

 2  1
f s3 s3    g s3 s
 
i 1  i 1  (4.23)
 e 
  i  1  1   2  0
3 3 3
 i

i 1  i 1 

1 1  2 2
 s3  1  u j  
2

 u j 1u2   3
 i 1u11 
  3 
  2  1 
1  u j du1du 2 
 
 ei  2  j 1 i 1
 2
1
u j
 i j 

 
j i 
B3 2,2,2 j 1
 0 0 

The next step, suppose that equation (4.1) holds for n  k , Then, it will be proven true for
n  k  1, such as W1 ,W2 ,...,Wk 1 are k  1 random variables where Wi with i  1,..., k  1 has
the probability density function f Wi , that has gamma distribution so that

 i2
f Wi wi   wi e  w  i i

e  i   i  1  1 (4.24)

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

It is assumed that S k , and Wk 1 , are identically independent distributed with

S k 1  S k  Wk 1 , Moreover, it shows that S k 1 has the probability density function f s k for
each s  R  with the convolution

f sk 1 s k 1   f S k  f Wk 1 s k 1  (4.25)

Let the probability density function f s k for n  k be true. Then, the probability density
function f s k is obtained as follows

 i
2  k 

 2  1
f sk s k  
 
i 1  i 1 
s (4.26)
 e   i  1  1   2  
k k k
 i

i 1  i 1 

1 1  j 
  2  1 
 u j 1  u j du1 ...du k 1 
  k  
 ... e k 1 n 1, ..., k 1
       1  i j 

 0 0
s P ,..., u u 1 u k k 1

Bk 1 2,... ,2 j 1
 

With
k 1 k 1 k 1
P 1 ,...  n u1,...uk 1  : 1  u j    i 1  u11  u j (4.27)
j 1 i 2 j i

It will be shown that the probability density function f sk 1 is also true where for the random
variable with the sum of n  k  1, it has the following probability density function

k 1

 i
2  k 1 

 2  1
f sk 1 s k 1  
  (4.28)
i 1  i 1 
s
 e    
k 1
k 1
 k 1
 i
i  1  1   2 
i 1  i 1 

1 1  j 
  2  1 
 sk 1 P 1 ,... n u1,...u k 1   1u k  k 1
k 1

 u j 1  u j du1 ...du k 1 
 
 ... e 1  i j 

 0 0 Bk 1 2,... ,2 j 1
 

Suppose that g is a continuous function on the real number line with

S k 1  W1  W2  ...  Wk 1 or it can be written S k 1  S k  Wk 1 . The following integral will be
calculated as

(4.29)
Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

I k 1 (g)  Eg s k  wk 1 ,

and it is defined as follows

k

1 1  i
2
1 1
 k 

 2  1 (4.30)
I k 1 g     g s k  wk 1    g sk  wk 1 s
 
i 1  i 1 

 e    
 
k k k
0 0  i
i  1  1   2  0 0
i 1  i 1 

1 1  1 
  2  1
 k 1 
  2  1 
 s k P 1 ,... n u1,...u k 1   k 1  1u k 1 
   
 ... e 1
1  u1 ...1  u k 1 du1 ...du k 1 
 i j   i j 

 0 0
u1 ....u k
Bk 1 2,... ,2
 

 k21
wk 1e  wk 1 k 1 ds k dwk 1
e  
 i
i  1  1 2 
k 1

 2  k 

 2 2  1

g y 
i 1 1
 

 
4 i 1 2
 y 22 y  i 1 
y 23 (4.31)

 e    
k 1
 k 1 1 1
 i
i  1  1   2  0 0
i 1  i 1 

1 1 2  1 
  2  1
 k 1 
  2  1 
 y1 P 1 ,... n u1,...u k 1    k 1 u k 1  y 22  k 1
   
 ... e 1
1  u1 ...1  u k 1 du1 ...du k 1 
 i j   i j 

 0 0
u1 ....u k
Bk 1 2,... ,2
 

dy1 dy 2

Furthermore, the spherical coordinates in equation (4.8) can be used to solve the equation
(4.31), so that the equation (4.31) becomes
k 1

 2   k 
 2  2  1  k 
 2  2  1

  g r sin  k  cos  k r  k r cos 3  k 

i 1 2  

I k 1 g   4
 i 1   
 i 1 
i 1 2 2 2
sin
 e     2
k 1 k 1
 i
i  1  1 0 0

i 1 i 1
(4.33)

1 1  1 
  2  1
 k 1 
  2  1 
 r 2 sin2  k P 1 ,... n u1,...u k 1   k 1u k 1  r 2 cos 2  k  k 1
   
 ... e 1  u1 ...1  u k 1 du1 ...du k 1  d k dr
 i j   i j 

 0 0
u1 ....u k
 

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

Given S k 1  r 2 and u j  sin 2  j for all j  1,... , n  1 , so that the equation (4.33) can be
written in the form of the following equation
k 1

 i
2
1 1
 k 
  2  1
 

I k 1 g     g s  s
 i 1 
i 1
(4.35)
 e    2
k 1 k 1 k 1 k 1
 i
i  1  1 0 0

i 1 i 1

1 1  1 
  2  1
 k 1 
  2  1 
  sk 1wk P 1 ,... n u1,...u k 1   k 1u k 1  s k 1 1u k  k 1
   
 ... e 1
1  u1 ...1  u k du1 ...d k  du k ds k 1
 i j   i j 

 0 0
u1 ....u k
Bk 1 2,...,2
 

or equation (4.35) can be written as

k 1

 i
2
1 1
 k 
  2  1
 

  g s  s
 i 1 
f s k 1  i 1

 e    2
k 1 k 1 k 1 k 1 (4.36)
 i
i  1  1 0 0

i 1 i 1

  k 
  2  1
  
 1 1 sk 1  1  u j   i 1u11   u j 1un 1  n 
 1 k k   i j 

 
k
1
  ... e
0 0
 j 1 i2 j i 
 j
Bk 1 2,...,2 j 1
u 1  u j du1 ...d k


 

By obtaining the equation (4.36), then the equation (4.1) holds for the case of n  k  1Based
on the mathematical induction, theorem (4.1) has been proven.

5. Convolution of Gamma Distribution with a Stabilizer Constant for Equal Parameters

Convolution of Gamma distribution with a stabilizer constant for the equal parameters is
given as follows.

Proposition 5.1. Suppose that there are n identically independent variables namely Wi for
i  1,... , n such that Wi has the density probability function f wi which is defined as follows
2
f W wi     wi e  w  i

e    1  1
i (5.1)

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

for 0  w  1, with i  1,2, ..., n and   0 . Then the sum of the random variables Wi , that is
n
S n   Wi which has the following probability density function
i 1
n

n 2 n

   i 1  2

f S n s n     
1 i 1
 s e   sn (5.2)
 e    1  1   2
n n

 i 1

Proof. This theorem will be proven by direct proof. The first step in proving this theorem is
to assume g as a continuous function on the real number line. Then, integral I n (g ) will be
calculated as

I n (g)  Eg w1  ...  wn , (5.3)

In which it can be defined as follows
 
I n (g)   ....  g w1  ...  wn  f w1 ... f wn dw1 ... dwn (5.4)
 
n n 
   2 
 ...  g w1  ...  wn w1 .... wn e 1 n dw1 ...dwn
1    w ...  w 
      
 e    1  1   2  0 0

Moreover, the variable transformation is done by assuming wi  yi where

2

0  yi  1 for i  1,2,...., n so that equation (5.4) becomes

 
n n
   2   
  ...  g y1 2  ...  y n 2 y13 .... y n 3 e    y1 ...  y n dy1 ...dy n (5.5)
1
I n ( g )  2     
2 2
n

 e    1  1   2  0 0
In addition, by using spherical coordinates in equation (4.8), it gives the result

n n
 1   2  n 1 2 n 1 n 1
I n ( g )  2   
n

    
 
  
    sin 3k
 k cos  k  sin k 1  k d k
3k
(5.6)
 e 1 1   2  k 1 0 k 1 k 1

 g (r )r e dr
2 4 n 1   r
2

 1  2
n 1
  n

      g (r 2 )r 4 n1e   r dr
2

 e    1  1   2  0
n


 i 1 

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

Next, the variable transformation of s n  r 2 is done in equation (5.6) above such that it yields
the following equation
 1
n
 2   n
2 r
4n
2r

I n ( g )      g ( r ) e   r dr
2
(5.7)
 e    1  1   2  0
n
r 2r

 i 1 


  g ( s n )  

1  2

n
  n
2 n 1
s n e   sn d s n
 e    1  1   2 
n
0

 i 1 

Based on equation (5.7) above, it shows that the random variable S n has the probability
density function f S n as follows
n

n 2 n

   i 1  2

f S n s n     
1 i 1
 s n e   sn (5.8)
 e    1  1   2
n

 i 1

Convolution of random variables X i for i  1,2,..., n which has gamma distribution, with
using the same parameter  for each random variable X i which has the probability density
function of f S n like equation (5.8) above.

6. Characteristics of the Convulution on Gamma Distribution with a Stabilizer Constant

for Equal Parameters

Characteristics of the convolution on Gamma Distribution with a Stabilizer Constant for

the equal parameters are as follows

Proposition 6.1. If W ~ Gammastabilizer(2,  ) with   2 and   0 then its cumulative

distribution function can be written as follows
 2, w
FW w;2,      , if w  0 (6.1)
e    1  1
Proposition 6.2. Let W be a random variable with a distribution of W ~ Gammastabilizer(2,  )
with   2 and   0 then

EW  
2
, and Var (w) 

6 e      1  1  4 
 e    1  1
(6.2)
 e    1  1
 2  2

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar
 The Convolution of Gamma Distribution with A Stabilizer Constant

Proposition 6.3. If W ~ Gammastabilizer(2,  ) then its moment generating function can be

written as follows
 2 (t   ) 2
M w (t )  (6.3)
e   (   1)  1

References

[1]. Akkouchi, Mohamed. 2008. On The Convolution of Eksponensial Distributions. Journal

of the Chungcheong Mathematical Societ, Vol. 21, No.4.
[2]. Anas, Marnisyah. 2013. Konvolusi Distribusi Eksponensial. Tesis Magister Matematika
Universitas Andalas, Padang
[3]. Bain, L.J, dan M. Engelhardt. 1992. Introduction to Probability and Mathematical
Statistic Second Edition. Duxbury Press, California.
[4]. Barnet, S dan R. G, Cameron. 1985. Introduction to Mathematical Control Theory
Second Edition. Clarendon Press. Oxford.
[5]. Casella, G dan R. L, Berger. 1990. Statistical inference. Ed ke-1, Pasific Grove,
California.
[6]. Devianto, Dodi. Maiyastri. Oktasari Lutfiana dan Marnisyah Anas. 2015. Convolution of
Generate Random Variable from Exponential Distribution with Stabilizer Constant.
Applied Mathematical Sciences, Vol. 9, 4781-4789.
[7]. Gnedenk, B. V, dan A. N. Kolmogorov. 1968. Limit Distribution Sums of Independent
Random Variabl. 2nd ed, Addison-Wesley, London.
[8]. Laha, R. G dan V. K. Rothatgi. 1979. Probability Theory. Jhon Wiley and Sons. New
York.
[9]. Rao, M. M, dan R. J. Swift. 2006. Probability with Applications Second Edition.
Springer, United States of America.
[10]. Rosita. 2012. Kajian Estimasi Distribusi Gamma untuk Mengatasi Kehilangan
Data Penjualan Pulsa. Matematika FMIPA-ITS, Surabaya.
[11]. Tucker, H. G. 1967. Probability and Mathematical Statistic. Academic Press.
London.

Rika Okda Marlina Zen, Radhiatul Husna, Dodi Devianto, Ferra Yanuar