EUROPEAN JOURNAL OF PURE AND APPLIED MATHEMATICS

Vol. 16, No. 1, 2023, 97-111

ISSN 1307-5543 – ejpam.com
Published by New York Business Global

Moment Generating Function of Current Records

Based on Generalized Exponential Distribution with
Some Recurrence Relations
Ramy Abdelhamid Aldallal
Department of Accounting, College of Business Administration in Hawtat bani Tamim,
Prince Sattam bin Abdulaziz University, Saudi Arabia

Abstract. This paper tries to find some formulas for calculating the moment generating function
for upper and lower current records picked from generalized exponential distributed data and the
joint moment generating function between them. After that, some formulas are derived from the
previous ones to find the moments of each and the product moments of both upper and lower
current records. Then, various recurrence relations are established for most of the mentioned
formulas. After that, an integral form of the moments of record range is founded followed by a
numerical example with simulated data to clarify the effectiveness of the formulas found in the
study and how they can make the calculation process easier and faster. Finally, a conclusion part
is added, to sum up what has been done and the results.
2020 Mathematics Subject Classifications: 62G30, 65Q30, 62E99, 60E10
Key Words and Phrases: Record range, lower and upper current records, moment generating
function and moments, recurrence relations, generalized exponential distribution

1. Introduction

Imagine that we have a sequence {Xj } that consists of identical and indepen-
dent random variables such that each of which is distributed with a probability density
function(pdf) f (x) and a cumulative distribution function(cdf) F (x) that is absolutely
continuous. Xj as an observation is considered to be an upper record if Xj > Xi for
every i < j. A similar definition has been established for the lower record values. In some
situations, we tend to pick the smallest and largest X values detected as new lower or
upper record values of either kind take place, and in such a situation, we name it cur-
rent records. We will symbolize Unc as the nth upper current records and Lcn as the nth
lower current records of the sequence Xn , when any kind o the nth records appears. So,
c
Un+1 = Unc if Lcn+1 < Lcn and Lcn+1 = Lcn if Un+1c > Unc , for all n = 1, 2, ... where by
definition, L0 = U0 = X1 . We can define the record range as Rnc = Unc − Lcn . And of
c c

DOI: https://doi.org/10.29020/nybg.ejpam.v16i1.4641
Email address: [email protected] (R.A.Aldallal)

https://www.ejpam.com 97 © 2023 EJPAM All rights reserved.

R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 98

course, a new range is recorded at the moment a new upper or lower record has occurred.
For example, let’s consider the following sequence of observations:

3, 2, 2.5, 2.2, 1, 3.7, 2.6, 1.5, 2.7, 4, 0.5, 2.5, 4.7....

Then we can determine the upper and lower current records along with the current record
range as follows:
U0c = 3, U1c = 3, U2c = 3, U3c = 3.7, U4c = 4, U5c = 4, U6c = 4.7
Lc0 = 3, Lc1 = 2, Lc2 = 1, Lc3 = 1, Lc4 = 1, Lc5 = 0.5, Lc6 = 0.5
R0c = 0, R1c = 1, R2c = 2, R3c = 2.7, R4c = 3, R5c = 3.5, R6c = 4.2
There are cases in which it is interesting for the current records to be considered in
real life. Such as whether data where lower and upper records are being taken together.
Also, in the case of outlier detection and when choosing a fitted model in which the record
range plays an important role in it (see, Basak [10]). Again, it is very substantial to be
used when we want to see if the production process is good enough to fall within the scope
of the production’s specifications or not. Meaning, if the record range is greater than a
specific value, then the products will not fit the specifications and vice versa. And you can
see it also in any life test in which we made sequential measurements and recorded only
the values that fall below or above a current extreme value like what happens in industrial
stress.
Barakat et al. [8] obtained some current record recurrence relations for some distribu-
tions and also moments recurrence relations for record range when the data follows the
exponential distribution. Once again, Barakat et al. [9] worked on the current record and
record range, but at this time, they established a prediction interval for a future value
of them. To read more about the record range and current records and the applications
titled to them, we pointed to Raqab[20], Ahmadi et al.[3] and Ahmadi and Balakrishnan[1]
and [2]. But to read about the record values themselves, you can check Aldallal [4], Abd
Elgawad et al. [12], Amany et al. [5], Husseiny et al.[17] and Barakat and Harpy [7].
Some of the data in the lifetime examples mentioned above can be distributed using
a distribution introduced by Gupta and Kundu[13] called the generalized exponential
distribution (GED) which can be used as a good alternative for gamma or the Weibull
model (see, Gupta and Kundu([13],[15], [14] and Mohie El-Din and Sharawy [11]). They
also mentioned that the GED has more similarities to a gamma family than a Weibull
family regarding the hazard function.
In this paper, we will introduce some new moment generating function (MGF) formu-
las for Unc , Lcn and the joint MGF between (Unc and Lcn ) based on GED along with their
ith moments. Also, some recurrence relations are introduced for the MGF of them beside
the recurrence relation of their ith moments. Moreover, an integrated form for the mth
moments for the record range also follows a GED has been driven. These results can help
minimize the direct computation of these MGFs and moments since numerical steps will
be applied.
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 99

2. Preliminaries Results

Houchens [16] concluded the pdf of Unc , Lcn , Rnc , and the joint pdf between (Unc
and Lcn ) when if follows any cdf F (x) respectively by

n n−1
X [− log F (x)]k o
n
fLcn (x) = 2 f (x) 1 − F (x) , (1)
k!
k=0

n n−1
X [− log F̄ (x)]k o
fUnc (x) = 2n f (x) 1 − F̄ (x) , (2)
k!
k=0
∞
2n
Z
fRnc (r) = f (r + x)f (x){− log[1 + F (x) − F (r + x)]}n−1 dx, 0 < r < ∞, (3)
(n − 1)! −∞
2n n on−1
fLcn ,Unc (x, y) = f (x)f (y) − log[1 + F (x) − F (y)] , (4)
(n − 1)!
where F̄ (x) = 1 − F (x).
Gupta and Kundu[13] established GED with the following pdf

f (x; α, λ) = αλ(1 − e−λx )α−1 e−λx , (5)

and the coming cdf

F (x) = (1 − e−λx )α . (6)
Where α is a shape parameter, λ is a scale parameter and a location parameter µ = 0.

Remark 1. The followings are some previously introduced expansions:

1. The logarithmic expansion introduced by Balakrishnan and cohen [6]
∞ i ∞
tp
X X
i
[−ln(1 − t)] = = ap (i)ti+p , |t| < 1 (7)
p
p=1 p=0

P i
∞ tp
where ap (i) is the coefficient of ti+p in the expansion p=1 p .
2. We have
βx (a, b)
Ix (a, b) = ,
β(a, b)
where Ix (a, b),βx (a, b) and β(a, b) represents the regularized incomplete beta function, the
incomplete beta function and the beta function respectively. We can find many series
representations for Ix (a, b) in many books and journals like the following which has been
introduced by Pearson [18]
a−1  
a+b−1
X a + b − 1  x i
Ix (a, b) = 1 − (1 − x) ,
i 1−x
i=0
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 100

from which we conclude that

a−1  
h
a+b−1
X a + b − 1  x i i
βx (a, b) = β(a, b) 1 − (1 − x) . (8)
i 1−x
i=0

Remark 2. We can conclude the following characterized differential equation

1 λx
F (x) = (e − 1)f (x), (9)
λα
which will be used in some of the recurrence relations. And this can be done by replacing
the term (1 − e−λx )α from (5) by F (x), which will lead to

e−λx
f (x) = αλF (x) .
1 − e−λx
Then
1 1 − e−λx 1 1
F (x) = −λx
f (x) = ( −λx − 1)f (x) = .
αλ e αλ e
Remark 3. The following relations will be used in the proof of some theorems (see, Raqab
[19])
βx (a, b) = a−1 xa 2 F1 (a, 1 − b; a + 1; x), (10)
Z 1
ua−1 (1 − u)b−1 2 F1 (c, d; ρ; u)du = β(a, b) 3 F2 (a, c, d; ρ, a + b; 1). (11)
0
Where 2 F1 and 3 F2 are generalized hypergeometric function’s special cases.

3. Moment Generating Functions

In this section, we will introduce some new formulas for calculating the MGF of
the lower current record MLcn (t), MGF of the upper current record MUnc (t) and joint MGF
of lower and upper current records MLcn ,Unc (t1 , t2 ) all based on GED.

Theorem 1. For n ≥ 2
n−1
X αk X ∞
h t t t i
MLcn (t) = 2n α β(α, 1 − ) − β(2α, 1 − ) − ap (k)β(2α, 1 + k + p − ) .
λ λ k! λ
k=1 p=0
(12)
While
n−1 ∞
n
h t X 1 X n t t oi
M Unc (t) = 2 α β(2α, 1 − ) − ap (k) β((1 + k + p)α, 1 − ) − β((2 + k + p)α, 1 − ) .
λ k! λ λ
k=1 p=0
(13)
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 101

And
∞ n+p−1
2n α 2 X X n + p − 1 t1
MLcn ,Unc (t1 , t2 ) = ap (n − 1) (−1)k β(αk + 1, 1 − )
(n − 1)! k λ
p=0 k=0
(14)
αk 
αk − tλ1 + 1

h t2 X t1 t2 i
β(α, 1 − )− β((n + p − k)α − 1 + i, 2 + i + αk − − ) .
λ i λ λ
i=0

Proof. By substituting (5) and (6) in (1), we get

n n−1
X [− log(1 − e−λx )α ]k o
n −λx α−1 −λx −λx α
fLcn (x) = 2 αλ(1 − e ) e 1 − (1 − e )
k!
k=0
n
= 2n αλ(1 − e−λx )α−1 e−λx 1 − (1 − e−λx )α − (1 − e−λx )α
n−1
X [− log(1 − e−λx )α ]k o
.
k!
k=1

So
Z ∞ Z ∞
−(λ−t)x −λx α−1
M Lcn
n
(t) = 2 αλ e (1 − e ) n
dx − 2 αλ e−(λ−t)x (1 − e−λx )2α−1 dx
0 0
∞ n−1
αk [− log(1 − e−λx )]k
Z X
−2n αλ e−(λ−t)x (1 − e−λx )2α−1 dx
0 k!
k=1
t t
= 2n α[β(α, 1 − ) − β(2α, 1 − )] − K(t).
λ λ
Upon using the substitution a = e−λx in K(t) and then applying the logarithmic expansion
from (7), we reach the following result
n−1 ∞
X αk+1 X t
K(t) = 2n ap (k)β(2α, 1 + k + p − ),
k! λ
k=1 p=0

which completes the proof of (12). Now to proof (13), we will substitute (5) and (6) in
(2) and with a routine calculation we get

n n−1
X [− log(1 − (1 − e−λx )α )]k
fUnc (x) = 2n αλ(1 − e−λx )α−1 e−λx − + (1 − e−λx )α
k!
k=1
n−1
X [− log(1 − (1 − e−λx )α )]k o
+(1 − e−λx )α .
k!
k=1
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 102

Then
Z ∞ Z ∞
−(λ−t)x −λx 2α−1
MUnc (t) = 2 αλ n
e (1 − e ) dx − 2 αλ n
e−(λ−t)x (1 − e−λx )α−1
0 0
n−1 ∞
[− log(1 − (1 − e−λx )α ]k
X Z
dx + 2n αλ e−(λ−t)x (1 − e−λx )2α−1
k! 0
k=1
n−1
X [− log(1 − (1 − e−λx )α ]k t
dx = 2n αβ(2α, 1 − ) − V (t) + Z(t).
k! λ
k=1

By applying the substitution b = (1 − e−λx )α then using the logarithmic expansion (7) for
both V (t) and Z(t) we reach (13). To prove (14), we will apply (5) and (6) into (4) and
this will lead to
2n
fLcn ,Unc (x, y) = α2 λ2 (1 − e−λx )α−1 e−λx (1 − e−λy )α−1 e−λy
(n − 1)!
n on−1
− log[1 − (1 − e−λy )α + (1 − e−λx )α ] .

And
∞
2n α 2 λ 2
Z
MLcn ,Unc (t1 , t2 ) = e(t2 −λ)y (1 − e−λy )α−1 I(y)dy,
(n − 1)! 0
where
Z y n on−1
I(y) = e(t1 −λ)x (1 − e−λx )α−1 − log[1 − (1 − e−λy )α + (1 − e−λx )α ] dx.
0

Let u = (1 − e−λy )α − (1 − e−λx )α , then using (7) we get

∞ Z (1−e−λy )α o −t1
1 X n 1
un+p−1 1 − [(1 − e−λy )α − u] α
λ
I(y) = ap (n − 1) du,
αλ 0
p=0
1
again using another substitution w = [(1 − e−λy )α − u] α then a binomial expansion on one
of the terms, we get
∞ n+p−1
X n + p − 1
1X
I(y) = ap (n − 1) (−1)k (1 − e−λy )α(n+p−1−k) J(y),
λ k
p=0 k=0

where Z 1−e−λy −t1

J(y) = (1 − w) λ wαk dw.
0
By applying (8) on J(y), we get
∞ n+p−1
X n + p − 1
1X −t1
I(y) = ap (n − 1) (−1)k (1 − e−λy )α(n+p−1−k) β(αk + 1, + 1)
λ k λ
p=0 k=0
αk  t1
+ 1 (1 − e−λy )i i

h
−λy αk−
t1
+1
X αk − λ
1 − (e ) λ .
i (e−λy )i
i=0
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 103

Now
∞ n+p−1
2n α2 λ X X n + p − 1
MLcn ,Unc (t1 , t2 ) = ap (n − 1) (−1)k (1 − e−λy )α(n+p−1−k)
(n − 1)! k
p=0 k=0
αk 
αk − tλ1 + 1

−t1 h X i
β(αk + 1, + 1) Q(t2 ) − R(t2 ) ,
λ i
i=0

where Z ∞
Q(t2 ) = e(t2 −λ)y (1 − e−λy )α(n+p−k)−1 dy,
0
and Z ∞ t1
R(t2 ) = et2 y (e−λy )2+i+αk− λ (1 − e−λy )α(n+p−k)−1+i dy.
0

Substituting z = e−λy
in Q(t2 ) and R(t2 ), and with some routine calculations, we reach
(14). And this completes the proof.

Corollary 1. The following recurrence relation evaluated from recursively differentiating

β(α, 1 − λt ) with respect to t for (i) times

i−1  
t i−k−1 1 i−k i − 1 t h t t i
X
(i)
β (α, 1− ) = (−1) ( ) β (k) (α, 1− ) ψ (i−k−1) (α+1− )−ψ (i−k−1) (1− ) .
λ λ j λ λ λ
k=0
(15)
Where β (i) (α, 1 − λt ) is the ith derivative of β(α, 1 − λt ) and ψ (i) (α) is the ith derivative of
the Poly-gamma function ψ(α).

Corollary 2. By differentiating (12) and (13) i times and substituting t = 0, we get

(i) (i)
the ith moment of lower current record µLcn and ith moment of upper current record µUnc
respectively. Also, by differentiating (14) j times with respect to t2 and i times with respect
to t1 and then substituting t1 = t2 = 0 we get the product moment of the lower and upper
(i,j)
current record µLcn ,Unc .

n−1
X αk X ∞
(i)
h t t t i
µLcn = 2n α β (i) (α, 1 − ) − β (i) (2α, 1 − ) − ap (k)β (i) (2α, 1 + k + p − ) .
λ λ k! λ t=0
k=1 p=0
(16)
While
n−1 ∞
(i) n
h
(i) t X 1 X n t
µUnc = 2 α β (2α, 1 − ) − ap (k) β (i) ((1 + k + p)α, 1 − )
λ k! λ
k=1 p=0 (17)
t oi
−β (i) ((2 + k + p)α, 1 − ) .
λ t=0
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 104

4. Some Recurrence Relations

Here we tried to find recurrence relations for MLcn+1 (t) and MUn+1
c (t) based on
previous terms MLcn (t) and MLcn (t) along with some additional terms. Also, the same has
(i) (i)
been done for µLc and µU c . Again, the same has been done for MLcn+1 ,Un+1 c (t1 , t2 )
n+1 n+1
based on previous function MLcn ,Unc (t1 , t2 ) with some added terms and for the product
(i,j)
moments µLc c .
n+1 ,Un+1

Theorem 2. For n ≥ 2
∞
(2α)n+1 X t
MLcn+1 (t) = 2MLcn (t) − ap (n)β(2α, − + n + p + 1) (18)
n! λ
p=0

∞
2n+1 α X h t t i
M c
Un+1 (t) = 2MUnc (t)− ap (n) β((1+n+p)α, 1− )−β((2+n+p)α, 1− ) (19)
n! λ λ
p=0

Proof. By replacing n by n + 1 in (12) and (13), and by some routine calculations we

reach (18) and (19) respectively.

Theorem 3. Knowing MLcn ,Unc (t1 , t2 ) and for n ≥ 2, we can calculate MLcn+1 ,Un+1
c (t1 , t2 )
using the following
∞ ∞ n+p+k−1
2n+1 X
 
n X
2
X
ω n+p+k−1
MLcn+1 ,Un+1
c (t1 , t2 ) = 2MLcn ,Unc (t1 , t2 ) + ap (n) α (−1)
(n − 1)! ω
p=0 k=0 ω=0

h α(n+p+k−ω+1)−1  
i α(n + p + k − ω + 1) − 1 t2
X
(−1) β(α(ω + 1) + 1, i − + 1)
i λ
i=0
 t1 t2 
F
3 2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + i − + 2; 1
λ λ
α(n+p+k−ω)−1  
X α(n + p + k − ω) − 1 t2
− (−1)l β(α(ω + 1) + 1, l − + 1)
l λ
l=0
 t1 t2 i
3 F2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + l − + 2; 1
λ λ
p+n  
t1 X n+p −t2
+ (−1)h β(α(p + n + 2), + 1)
nλ(λ + 1) h λ
h=0
h  t1 t2 
3 F2 α(p + n + 2)), α(h + 1), 1 + ; α(h + 1) + 1, α(p + n + 2) + 1 − ; 1
λ λ
 t1 t2 io
− 3 F2 α(p + n + 2)), α(h + 1), ; α(h + 1) + 1, α(p + n + 2) + 1 − ; 1 .
λ λ
(20)
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 105

Proof. We already know that

2n+1 ∞ y t1 x+t2 y
Z Z
MLcn+1 ,Un+1
c (t1 , t2 ) = e f (x)f (y)− log[1 + F (x) − F (y)]n dxdy.
n! 0 0
Let Z y
I(y) = et1 x f (x){− log[1 + F (x) − F (y)]}n dx,
0
using integration by parts by putting the deferential part to be et1 x {− log[1+F (x)−F (y)]}n
and the rest as the integral part we reach
Z ∞Z y
2n+1 F (x)
MLcn+1 ,Un+1
c (t1 , t2 ) = et1 x+t2 y f (x)f (y)
(n − 1)! 0 0 1 + F (x) − F (y)
Z ∞Z y
n−1 2n+1
{− log[1 + F (x) − F (y)]} dxdy − t1 et1 x+t2 y f (y)F (x)
n! 0 0
{− log[1 + F (x) − F (y)]}n dxdy,
from which, we can simply conclude that it will became
Z ∞Z y
2n+1 F (y) − 1
MLcn+1 ,Un+1c (t1 , t2 ) = 2MLcn ,Unc (t1 , t2 ) + et1 x+t2 y f (x)f (y)
(n − 1)! 0 0 1 + F (x) − F (y)
Z ∞Z y
n−1 2n+1
{− log[1 + F (x) − F (y)]} dxdy − t1 et1 x+t2 y f (y)F (x)
n! 0 0
{− log[1 + F (x) − F (y)]}n dxdy.
(21)
Now choose
Z ∞Z y
F (y) − 1
J(t1 , t2 ) = et1 x+t2 y f (x)f (y){− log[1 + F (x) − F (y)]}n−1 dxdy,
0 0 1 + F (x) − F (y)
and Z ∞Z y
ζ(t1 , t2 ) = et1 x+t2 y f (y)F (x){− log[1 + F (x) − F (y)]}n dxdy.
0 0
From J(t1 , t2 ), let
Z ∞
f (x)
K(y) = et1 x {− log[1 + F (x) − F (y)]}n−1 dx,
0 1 + F (x) − F (y)
now, substituting u = F (y) − F (x) and then using the logarithmic expansion from (7) we
reach
∞ Z F (y) n−1+p
X u 1 −t1
K(y) = ap (n) [1 − (F (y) − u) α ] λ du,
0 1−u
p=0
1
= ∞ k
P
since −1 < u < 1, we can use the well known expansion 1−u k=0 u , then make a
1
substitution of w = (F (y) − u) α to reach
∞ ∞ n+p+k−1  
ω n+p+k−1 t1
X X X
K(y) = α ap (n) (−1) (F (y))n+p+k−ω−1 β 1 (α(ω + 1), 1 − ).
ω (F (y)) α λ
p=0 k=0 ω=0
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 106

Now J(t1 , t2 ) has became

∞ ∞ n+p+k−1  Z ∞
ω n+p+k−1
X X X
J(t1 , t2 ) = α ap (n) (−1) f (y)(F (y) − 1)(F (y))n+p+k−ω−1 et2 y
ω 0
p=0 k=0 ω=0
t1
β 1 (α(ω + 1), 1 − )dy.
(F (y)) α λ
By replacing f (y) and F (y) by their original formulas from (5) and (6) and after that we
made a substituting e−λy = u, we get
∞ ∞ n+p+k−1  h α(n+p+k−ω+1)−1
ω n+p+k−1
X X X X
2
J(t1 , t2 ) = α ap (n) (−1) (−1)i
ω
p=0 k=0 ω=0 i=0
 Z 1
α(n + p + k − ω + 1) − 1 t2 t1
ui− λ β1−u (α(ω + 1), 1 − )du
i 0 λ
α(n+p+k−ω)−1  Z 1
X α(n + p + k − ω) − 1 t2 t1 i
− (−1)l ul− λ β1−u (α(ω + 1), 1 − )du ,
l 0 λ
l=0

upon using (10) and (11) in the last integral we reach the final form of the function J(t1 , t2 )
as follows
∞ ∞ n+p+k−1  h α(n+p+k−ω+1)−1
ω n+p+k−1
X X X X
2
J(t1 , t2 ) = α ap (n) (−1) (−1)i
ω
p=0 k=0 ω=0 i=0
 
α(n + p + k − ω + 1) − 1 t2
β(α(ω + 1) + 1, i − + 1)
i λ
 t1 t2 
(22)
F
3 2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + i − + 2; 1
λ λ
α(n+p+k−ω)−1  
l α(n + p + k − ω) − 1 t2
X
− (−1) β(α(ω + 1) + 1, l − + 1)
l λ
l=0
 t1 t2 i
3 F2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + l − + 2; 1 .
λ λ
And for ζ(t1 , t2 ) we begin by taking the following integration out of it
Z y
Z(y) = et1 x F (x){− log[1 + F (x) − F (y)]}n dx,
0
then making a substitution of u = F (y) − F (x), after that we used equation (9) followed
1
by the logarithmic expansion from (7). Another substitution made by w = ((F (y) − u) α
and some routine calculations leads to the following
∞ q+n
X n + q 
1X n −t1
Z(y) = aq (n) (−1)h (F (y))q+n−h β 1 (α(h + 1), )−
λ h (F (y)) α λ
q=0 h=0
−t1 o
β 1 (α(h + 1), + 1) .
(F (y)) α λ
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 107

Now, we return to the original ζ(t1 , t2 ) double integral and put the final form of Z(y)
function in it. After that, we replace F (y) and f (y) as in (5) and (6) respectively and by
the substitution u = 1 − e−λy we reach an integral in which we can use (10) and (11) to
solve it and reach that final formula of ζ(t1 , t2 ) which is going to be
∞ q+n
X n + q 
−1 X −t2
ζ(t1 , t2 ) = aq (n) (−1)h β(α(q + n + 2), + 1)
λ(λ + 1) h λ
q=0 h=0
n  t1 t2  (23)
3 F2 α(q + n + 2)), α(h + 1), 1 + ; α(h + 1) + 1, α(q + n + 2) + 1 − ; 1
λ λ
 t1 t2 o
− 3 F2 α(q + n + 2)), α(h + 1), ; α(h + 1) + 1, α(q + n + 2) + 1 − ; 1 .
λ λ
Exchanging (22) and (23) by their equal parts from (21), we get (20).

Lemma 1. Another recurrence relation for of MLcn+1 ,Un+1 c (t1 , t2 ) can be found by using
analogs steps used in Theorem 3 but leads to the following relation
t1 t1 2n+1 2
(1 − )MLcn+1 ,Un+1
c (t1 , t2 ) + MLcn+1 ,Un+1
c (t1 + λ, t2 ) = 2MLcn ,Unc (t1 , t2 ) + α
αλ αλ (n − 1)!
∞ ∞ n+p+k−1  
X X X n+p+k−1
ap (n) (−1)ω
ω
p=0 k=0 ω=0

h α(n+p+k−ω+1)−1  
i α(n + p + k − ω + 1) − 1 t2
X
(−1) β(α(ω + 1) + 1, i − + 1)
i λ
i=0
 t1 t2 
3 F2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + i − + 2; 1
λ λ
α(n+p+k−ω)−1  
X α(n + p + k − ω) − 1 t2
− (−1)l β(α(ω + 1) + 1, l − + 1)
l λ
l=0
 t1 t2 i
F
3 2 α(ω + 1) + 1, α(ω + 1), 1 − ; α(ω + 1) + 1, α(ω + 1) + l − + 2; 1 .
λ λ
(24)
Corollary 3. For i = 1, 2, 3, ..., and n ≥ 2, by taking the ith derivative of (18) and (19)
and substituting t = 0 we reach the following relations. Also, by differentiating (20) i
times with respect to t1 and j times with respect to t2 and then substituting t1 = t2 = 0 we
get the recurrence relation of the product moment between lower and upper current record
(i,j)
µLc ,U c .
n+1 n+1

∞
(i) (i) (2α)n+1 X t
µLc = 2µLcn (t) − ap (n)β (i) (2α, − + n + p + 1) . (25)
n+1 n! λ t=0
p=0
∞
n+1 α X
(i) (i) 2 h t t i
µU c = 2µUnc (t) − ap (n) β (i) ((1 + n + p)α, 1 − ) − β (i) ((2 + n + p)α, 1 − ) .
n+1 n! λ λ t=0
p=0
(26)
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 108

5. Moments of Record Range

(m)
A calculable formula for the mth moments of record range µRnc will be provided
in the next theorem along with its proof. The formula contains an integrated part that
can be easily calculated using any mathematical package such as MATHEMATICA.

Theorem 4. For n > 2

α−1 α−1 ∞  Z ∞
2n α2 λ X X X α − 1 α − 1 (−1)v+w
 
(m)
µRnc = rm e−λ(w+1)r
(n − 1)! v w v+w+n+j+1 0 (27)
v=0 w=0 j=0

aj (n − 1, e−λr , α)dr.

Proof. By substituting (5) and (6) in (3), we get

2n α2 λe−λr
fRnc (r) = λI(r), (28)
(n − 1)!

where
Z ∞ n
I(r) = λ (e−λx )2 (1 − e−λx )α−1 (1 − e−λr e−λx )α−1 − log[1 − (1 − e−λr e−λx )α
0
on−1
+(1 − e−λx )α ] dx.

By making the substitution u = eλx and then using the binomial expansion on some of
the terms, we reach
α−1
X α−1
X   Z 1
α−1 α−1 n on−1
I(r) = (−1)v+w (e−λr )w uv+w+1 −log[1−(1−e−λr u)α +(1−u)α ] du.
v w 0
v=0 w=0

The following expansion was created in the same way Balakrishnan and cohen [6] did in
(7)
n on−1 X ∞
−λr α α
− log[1 − (1 − e u) + (1 − u) ] = aj (n − 1, e−λr , α)un−1+j
j=0

, where aj (n − 1, e−λr , α) is the coefficient of the expansion. Then

α−1
X α−1 ∞    Z 1
XX α−1 α−1 v+w −λr w −λr
I(r) = (−1) (e ) aj (n − 1, e , α) uv+w+1 un−1+j du
v w 0
v=0 w=0 j=0
α−1
X α−1 ∞ 
e−λwr
 
XX α−1 α−1
= (−1)v+w aj (n − 1, e−λr , α) .
v w v+w+n+j+1
v=0 w=0 j=0
(29)
Now, substitute (29) in (28) we get
R.A.Aldallal / Eur. J. Pure Appl. Math, 16 (1) (2023), 97-111 109

α−1 α−1 ∞ 
2n α 2 λ X X X α − 1 α − 1 e−λ(w+1)r
 
fRnc (r) = (−1)v+w aj (n−1, e−λr , α) .
(n − 1)! v w v+w+n+j+1
v=0 w=0 j=0
(30)
To find the moments of the record range we will use the well-known formula
Z ∞
(m)
µRnc = rm fRnc (r)dr. (31)
0
Substituting (30) in (31) we get reach (27).

6. Numerical Example

To prove the benefits and motivations of using the previous formulas, a numerical sim-
ulation study was conducted to illustrate that.

Simulation Study. Samples of size 20 from the pdf of the upper current records (2)
for GED cdf when α = 5 and λ = 2 and at different values of n = 2, 3, 4, 5 and 6 was
simulated using MATHEMATICA 12.0 and the true mean for each sample was calculated.
After that, we used formula (17) to estimate the value of the mean at the same values
of α, λ and n when i = 1. Then, from (23) we calculated the predicted mean at also the
same α, λ, n and i. As we can see from the results listed in Table 1, the true mean is very
close to the estimated mean and the predicted mean is even closer to the estimated mean
indicating the efficiency of the formulas created in this paper. Analogies work has been
done for the lower current records formulas and the same good results were found.
Note: Most of the results calculated from the formulas (16),(17),(18), and (19) were con-
ducted to a maximum of 2000 iterations. And after that, the results did not change.
Table 1: True, Estimate and Predicted Mean

Upper Current Record n=2 n=3 n=4 n=5 n=6

True Mean 1.73759 2.03974 2.32234 2.52788 2.78866
Estimated Mean 1.75027 2.02074 2.28374 2.54529 2.81821
Predicted Mean 2.02066 2.28373 2.5453 2.81863
Lower Current Record n=2 n=3 n=4 n=5 n=6
True Mean 0.649044 0.52386 0.474306 0.403313 0.326389
Estimated Mean 0.649046 0.535685 0.452289 0.387566 0.335597
Predicted Mean 0.535681 0.452289 0.387566 0.335597

7. Conclusion

It is difficult to calculate the MGF of the lower or the upper current record or the joint
MGF of the lower and upper current record especially when it follows a distribution such
REFERENCES 110

as GED. The author tries to solve this by finding the simplest formulas to achieve that.
Although, the author establishes some formulas to find the moments of the lower current
record, upper current record, record range, and product moments of the lower and upper
current record based on the same distribution. Also, the larger the value n grows, the
more time it takes and the more difficult it becomes to be calculated. And here comes
the role of the recurrence relation formulas created for the moments and MGFs in this
paper to make it easier and faster. To demonstrate that, a numerical example is added
and most of the formulas are used in it and the desired result occurred by finding that
the formulas of the MGF and the moments give values very close to the actual ones. And
also, the values that result from the recurrence relations are almost the same values as the
ones from its opposite MGF or moments but calculated faster.

Acknowledgements

This publication was supported by Deanship of Scientific Research, Prince Sattam Bin
Abdul-Aziz University, Alkharj, Saudi Arabia. Our special thanks for their encouragement
and supporting of this research.
Also, I would like to thank Prof. Haroon Barakat for his valuable ideas that were very
helpful in creating this paper.

References

[1] J. Ahmadi and N. Balakrishnan. Distribution-free confidence intervals for quantile

intervals based on current records. Statistics and probability letters, 75(3):190–202,
2005.

[2] J. Ahmadi and N. Balakrishnan. Preservation of some reliability properties by certain

record statistics. Statistics, 39(4):347–354, 2005.

[3] J. Ahmadi, M. Razmkhah, and N. Balakrishnan. Current k-records and their use in
distribution-free confidence intervals. Statistics and Probability Letters, 79(1):29–37,
2009.

[4] R. A. Aldallal. Best linear unbiased estimation and prediction of record values based
on kumaraswamy distributed data. Statistics, Optimization & Information Comput-
ing, 10(4):1250–1266, 2022.

[5] E. A. Amany, H. M. Barakat, and M. E. El-Adll. Prediction intervals of the record-

values process. Revstat - Statistical Journal, 17(3):401–427, 2019.

[6] N. Balakrishnan and A. C. Cohen. Order statistics and inference: estimation methods.
Academic Press, San Diego, 1991.

[7] H. M. Barakat and M. H. Harpy. Asymptotic behavior of the records of multivariate

random sequences in a norm sense. Math. Slovaca, 70(6):1457–1468, 2020.
REFERENCES 111

[8] H. M. Barakat, E. M. Nigm, and R. A. Aldallal. Current records and record range
with some applications. Journal of the Korean Statistical Society, 43(2):263–273,
2014.

[9] H. M. Barakat, E. M. Nigm, and R. A. Aldallal. Exact prediction intervals for future
current records and record range from any continuous distribution. SORT, 38(2):251–
270, 2014.

[10] P. Basak. An application of record range and some characterization results. In

Advances on Theoretical and Methodological A spects of Probability and Statistics,
pages 83–95. CRC Press, 2019.

[11] M. M. Mohie El-Din and A. M. Sharawy. Characterization for generalized exponential

distribution. Mathematical Sciences Letters, 10(1):15–21, 2021.

[12] M. A. Abd Elgawad, H. M. Barakat, and Y. Ting. Bivariate limit theorems for record
values based on random sample sizes. Sankhya A: The Indian Journal of Statistics,
82(1):50–67, 2020.

[13] R. D. Gupta and D. Kundu. Theory and methods: Generalized exponential distribu-
tions. Australian and New Zealand Journal of Statistics, 41(2):173–188, 1999.

[14] R. D. Gupta and D. Kundu. Exponentiated exponential family: an alternative to

gamma and weibull distributions. Biometrical Journal: Journal of Mathematical
Methods in Biosciences, 43(1):117–130, 2001.

[15] R. D. Gupta and D. Kundu. Generalized exponential distribution: different method

of estimations. Journal of Statistical Computation and Simulation, 69(4):315–337,
2001.

[16] R. L. Houchens. Record value theory and inference. PhD thesis, University of Cali-
fornia, Riverside, 1984.

[17] I. A. Husseiny, H. M. Barakat, G. M. Mansour, and M. A. Alawady. Information

measures in records and their concomitants arising from sarmanov family of bivariate
distributions. Journal of Computational and Applied Mathematics, 408:114–120, 2022.

[18] K. Pearson. Tables of the incomplete beta function. Cambridge University Press,
London, 1968.

[19] M. Z. Raqab. Generalized exponential distribution: Moments of order statistics.

Statistics, 38(1):29–41, 2004.

[20] M. Z. Raqab. Bounds on expectations of record range and record increment from
distributions with bounded support. J Inequalities Pure Appl Math, 8(1), 2007.