J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.

6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

Some approximations of the Bateman’s G−function

Mansour Mahmoud1 , Ahmed Talat2 and Hesham Moustafa3
1,3 Mansoura University, Faculty of Science, Mathematics Department, Mansoura 35516, Egypt.
2 Port Said University, Faculty of Science, Mathematics and Computer Sciences Department,
Port Said, Egypt.
1 [email protected]

2a
− t− [email protected]
3 [email protected]

.
Abstract
In the paper, we presented a family M (µ, x) of approximations of the Bateman function
G(x). The family M (µ, x) = G(x) for a certain µ whenever x is fixed and it presented
asymptotical approximation of the Bateman’s G−function as x → ∞. We studied the or-
der of convergence of the approximations M (µ, x) of the function G(x). Some properties
and bounds of the error are deduced. We presented new sharp double inequality of G(x)
with the upper and lower bounds M (1, x) and M ( e24−4 , x) (resp.). Also, we show that the
approximations M (µ, x) are better than the approximation x1 + 2x12 for any µ in an open
[ ]
subinterval of 1, e24−4 .

2010 Mathematics Subject Classification: 33B15, 26D15.

Key Words: Bateman function, digamma function, monotonicity, sharp inequality, ap-
proximation, error.

1 Introduction.
In 1953, Erdélyi [6] defined the Bateman’s G−function as
( ) (x)
x+1
G(x) = ψ −ψ , x≠ 0, −1, −2, ... (1)
2 2

where the digamma function ψ(x) is given by

d
ψ(x) = log Γ(x)
dx

1165 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

and Γ(x) is the ordinary gamma function defined by [3]

∫ ∞
Γ(x) = tx−1 e−t dt, x > 0.
0

The function G(x) is very useful in estimating and summing certain numerical and algebraic se-
ries [18]. For more details on bounding the function Γ(x) and its logarithmic derivatives ψ (n) (x),
please refer to the papers [2]-[5], [7]-[23] and plenty of references therein.

The function G(x) can be also defined by

2
G(x) = 2 F1 (1, x; 1 + x; −1),
x
where
∑
∞
(a1 )k ...(ar )k xk
r Fs (a1 , ..., ar ; b1 , ..., bs ; x) =
k=0
(b1 )k ...(bs )k k!
is the generalized hypergeometric series [1] defined for r, s ∈ N, aj ∈ C, bj ∈ C − {0, −1, −2, ...}
and the Pochhammer symbol (a)n is defined by

∏
n−1
Γ(a + n)
(a)0 = 1 and (a)n = (a + i) = , n ≥ 1.
i=0
Γ(a)

The function G(x) satisfies the functional equation [6]:

2
G(1 + x) = −G(x) + (2)
x
and it has the integral representation
∫ ∞
e−xt
G(x) = 2 dt, x>0 (3)
0 1 + e−t

which can be deduced from the following known integral representation of the digamma [3]
∫ ∞ ( −t )
e e−xt
ψ(x) = − dt, x > 0.
0 t 1 − e−t

Qiu and Vuorinen [24] deduced the inequality

1 4(1.5 − log 4) 1 1
+ 2
< G(x) < + 2 , x > 1/2. (4)
x x x 2x
Mahmoud and Agarwal [9] presented the following asymptotic formula for Bateman’s G-function

1 ∑ (22k − 1)B2k
∞
G(x) ∼ + , x→∞ (5)
x k=1 kx2k

1166 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

and they deduced the double inequality

1 1 1
< G(x) − < 2 , x>0 (6)
2x2 + 1.5 x 2x
which improve the lower bound of the inequality (4). Also, Mahmoud and Almuashi [11] proved
that the Bateman’s G−function satisfies the double inequality
∑
2m
(2n − 1)B2n 1 ∑ (2n − 1)B2n
2m−1
< G(x) − < , m∈N (7)
n=1
nx2n x n=1
nx 2n

with best bounds, where Br ′ s are the Bernoulli numbers and they presented some estimates for
the error term of a class of the alternating series, which improve and generalize some recent
results. Mortici [13] established the inequality
1
0 < ψ(x + v) − ψ(x) ≤ ψ(v) + γ + −v x ≥ 1; 0 < v < 1, (8)
v
where γ is the Euler constant, which also improves the inequality (4) of Qiu and Vuorinen. Also,
Alzer presented the double inequality [2]
1 1
− Tn (v; x) − ρn (v; x) < ψ(x + v) − ψ(x) < − Tn (v; x),
x x
where n ≥ 0 be an integer, x > 0, 0 < v < 1,
[ ]
1 ∑
n−1
1
Tn (v; x) = (1 − v) +
v + n + 1 i=0 (x + i + 1)(x + i + v)

and
1 (x + n)(x+n)(1−v) (x + n + 1)(x+n+1)v
ρn (v; x) = log .
x+n+v (x + n + v)x+n+v

In 2006, Muqattash and Yahdi [17] presented an infinite family of functions Ia (x) = ψ(x)
for a certain a when x is fixed. Local and global bounding error functions are found and new
inequalities for the Digamma function are introduced. These functions are shown to approximate
ψ locally and asymptotically. The approximations are compared to another approximations of
the Digamma function. The technique of construct of Muqattash and Yahdi is very useful and
can be updated to another functions as we will see in this paper.

In 2014, Guo and Qi improved the results of [8] and presented the two sharp inequalities
( )
1 1 ( )
ln x + < ψ(x) + < ln x + e−γ , x>0
2 x

where the constants 12 and e−γ are the best possible, and
( )
1 ( )
ln n + + γ < Hn (n) < ln n + e1−γ − 1 + γ, n∈N
2

1167 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

where the n-th harmonic numbers are defined by

∑
n
1
Hn = , n∈N
i=1
i

and is related to the Psi function by the relation

Hn = γ + ψ(n + 1).

In this paper, we presented a family of functions M (µ, x) satisfies that for all x > 0 there
exists µ ∈ [1, 2] such that M (µ, x) = G(x) and is asymptotically equivalent to G(x) as x → ∞.
We proved
( that the approximations
) M (µ, x) of the function G(x) are of an order of convergence
(x+2)[(e2 −4)x+4]
( )
of O ln (x+1)[(e2 −4)x+e2 ] for x > 2 and µ ∈ 1, e24−4 . Some properties and bounds of the error
are deduced. Also, we presented a new sharp double inequality of the function G(x) between
the lower bound M ( e24−4 , x) and the upper bound M (1, x). We proved that the approximations
[ ]
M (µ, x) are better than the approximation x1 + 2x12 for any µ in an open subinterval of 1, e24−4 .

2 Main Results
Lemma 2.1. For x > 0, we have
( ) ( )
1 2 1 2
ln 1 + + ≤ G(x) ≤ ln 1 + + . (9)
x+2 x(x + 1) x+1 x(x + 1)

Proof. Consider the function

( )
1 2
Hµ (x) = ln 1 + + − G(x), x > 0; µ > 0
x+µ x(x + 1)

which can be represented using (3) by the integral formula

∫ ∞
e−(µ+1)t [e2t − 1 − 2teµt ] −xt
Hµ (x) = e dt.
0 t(1 + et )

The function m1 (t) = e2t − 1 − 2tet is strictly increasing pass through the origin, then H1 (x) > 0,
that is ( )
1 2
ln 1 + + > G(x).
x+1 x(x + 1)
Also, m2 (t) = e2t −1−2te2t is strictly decreasing function pass through the origin, then H2 (x) < 0,
that is ( )
1 2
ln 1 + + < G(x).
x+2 x(x + 1)

1168 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

The double inequality (9) show that the function G(x) lies between two functions of the
following family of functions
( )
1 2
M (µ, x) = ln 1 + + x > 0; µ > 0. (10)
x+µ x(x + 1)

and hence we can conclude the following result:

Theorem 1. For every x > 0, there exists µ ∈ [1, 2] such that

M (µ, x) = G(x).

Proof. For a positive fixed x, consider the function M2 (µ) = M (µ, x) with 1 ≤ µ ≤ 2 and
G(x) = λ. M2 (µ) is a continuous on [1, 2] and using the inequality (9), we obtain

M2 (2) ≤ λ ≤ M2 (1).

Then by the Intermediate Value Theorem, there exists µ ∈ [1, 2] such that M2 (µ) = λ.
Also, by using the relations

∂M (µ, x) 2µ + 2µ2 + 2x + 8µx + 4µ2 x + 7x2 + 8µx2 + 6x3 + x4

=− <0
∂x x2 (1 + x)2 (µ + µ2 + x + 2µx + x2 )

and
∂M (µ, x) −1
= < 0,
∂µ (x + µ + 1)(x + µ)
we obtain the following properties of the family M (µ, x).

Lemma 2.2.

1. M1 (x) = M (µ, x) is a positive and strictly decreasing as a function of x, x > 0.

2. M2 (µ) = M (µ, x) is strictly decreasing as a function of µ, 1 ≤ µ ≤ 2

and hence
0 < M (2, x) ≤ M (µ, x) ≤ M (1, x), x > 0; µ ∈ [1, 2]. (11)

Now, we will show that the family M (µ, x) presented asymptotical approximation of the
Bateman’s G−function for all µ ∈ [1, 2].

Theorem 2. For all µ ∈ [1, 2], the Bateman’s G−function and the family M (µ, x) are asymp-
totically equivalent as x → ∞, that is

G(x)
lim =1
x→∞ M (µ, x)

and this is written symbolically as G(x) ∼ M (µ, x).

1169 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

Proof. Using the inequality (9), we get

M (2, x) ≤ G(x) ≤ M (1, x) (12)

and hence
M (2, x) G(x)
≤ ≤ 1.
M (1, x) M (1, x)
But
M (2, x) 12 + 34x + 23x2 + 6x3 + x4
lim = =1
x→∞ M (1, x) (3 + x)(4 + 10x + 5x2 + x3 )
and then
G(x)
lim = 1. (13)
x→∞ M (1, x)

Similarly, we have
G(x)
lim = 1. (14)
x→∞ M (2, x)
Using the inequality (11), we obtain

G(x) G(x) G(x)

≤ ≤ . (15)
M (1, x) M (µ, x) M (2, x)

From (13), (14) and (15), we get

G(x)
1 ≤ lim ≤ 1.
x→∞ M (µ, x)

Now, we will study the error of the approximation M (µ, x) of the function G(x).

Theorem 3. For any µ ∈ [1, 2], the error

eµ (x) = G(x) − M (µ, x)

approaches zero as x → ∞ and

( ) ( ( ))
1 2 1
G(x) = ln 1 + + + O ln 1 + . (16)
x+µ x(x + 1) (x + 1)(x + 3)

Proof. From inequality (12), we have

M (2, x) − M (µ, x) ≤ G(x) − M (µ, x) ≤ M (1, x) − M (µ, x)

and using (11), we get

M (2, x) − M (1, x) ≤ M (2, x) − M (µ, x).
Hence
0 ≤ |G(x) − M (µ, x)| ≤ M (1, x) − M (2, x) (17)

1170 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

or ( )
1
0 ≤ |eµ (x)| ≤ ln 1 + . (18)
(x + 1)(x + 3)
Then ( ( ))
1
G(x) = M (µ, x) + O ln 1 +
(x + 1)(x + 3)
and
lim eµ (x) = 0.
x→∞

As a consequence of the above result, we obtain some bounds of the error eµ (x).
( )
Corollary 2.3. The error eµ (x) is uniformly bounded by ± ln 1 + (ε+1)(ε+3)
1
∀ x > ε > 0 and
∀ µ ∈ [1, 2].
Proof. Using the inequality (18), we obtain
( )
1
sup |eµ (x)| ≤ ln 1 + .
0<x<∞ (x + 1)(x + 3)
( )
1
Also, the function g(x) = ln 1 + (x+1)(x+3) for x > 0 is decreasing. Then the errors eµ (x) are
( ) ( )
uniformly bounded between − ln 1 + (ε+1)(ε+3)
1 1
and ln 1 + (ε+1)(ε+3) .

3 The best bounds of the double inequality (9).

Firstly, we will prove the following auxiliary results:
Lemma 3.1. ( )
1
lim −x =1 (19)
x→∞ eG(x+2) − 1
and
G′ (x + 2)eG(x+2)
lim = −1. (20)
x→∞ (eG(x+2) − 1)2

Proof. Using the double inequality (6) with

1 1 1 1
β(x) = + 2 and α(x) = + 2,
x 2x + 3/2 x 2x
we get
( ) ( ) ( )
1 1 1
lim − x ≤ lim − x ≤ lim −x .
x→∞ eα(x+2) − 1 x→∞ eG(x+2) − 1 x→∞ eβ(x+2) − 1
But
( ) ( )
1 1
lim − x = lim [ ( 1 )] −x =1
x→∞ eα(x+2) − 1 x→∞ 1+ 1
x
− 1
x2
+ 2
3x3
5
+ 12x 4 − O x 5 − 1

1171 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

and
( ) ( )
1 1
lim − x = lim [ ( 1 )] − x = 1.
x→∞ eβ(x+2) − 1 x→∞ 1+ 1
x
− 1
x2
+ 2
3x3
1
+ 24x 4 − O x5
−1

Also, using the double inequality (6), we have

G′ (x + 2)eα(x+2) G′ (x + 2)eG(x+2) G′ (x + 2)eβ(x+2)

lim ≤ lim ≤ lim .
x→∞ (eβ(x+2) − 1)2 x→∞ (eG(x+2) − 1)2 x→∞ (eα(x+2) − 1)2

Now, using the asymptotic formula for Bateman’s G-function (5), we obtain
( )
′ −1 1
G (x) = 2 − O .
x x3

Then
[ ( )] [ ( 1 )]
−1
′
G (x + 2)e α(x+2)
(x+2)2
− O x13 1 + x1 − x12 + 3x23 + 12x5
4 − O x5
lim = lim [ ( 1 )] = −1
x→∞ (eβ(x+2) − 1)2 x→∞ ( 1 + x − x2 + 3x3 + 24x4 − O x5 − 1)2
1 1 2 1

and
[ ( 1 )] [ ( 1 )]
−1
G′ (x + 2)eβ(x+2) (x+2)2
− O x3 1 + x1 − x12 + 3x23 + 24x 1
4 − O x5
lim = lim [ ( )] = −1
x→∞ (eα(x+2) − 1)2 x→∞ ( 1 + x1 − x12 + 3x23 + 12x
5
4 − O
1
− 1)2
x5

Now, we will present the sharp bounds of the double inequality (9).

Theorem 4. For all x ∈ (0, ∞)

( ) ( )
1 2 1 2
ln 1 + 4 + < G(x) < ln 1 + + , (21)
x + e2 −4 x(x + 1) x+1 x(x + 1)

4
where the constants 1 and e2 −4
are the best possible.

Proof. Using the inequality (9) and functional equation (2), we get
1
0< − x < 2.
eG(x+2) −1
Now consider the two functions

f (x) = eG(x+2) − 1, x>0

and
1
q(x) = − x, x > 0.
f (x)

1172 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

Then f ′ (x) = G′ (x + 2)eG(x+2) < 0 and f (x) is strictly decreasing function. Hence f (x)
1
is strictly
increasing function. Since dx d 1
|
f (x) x=0
≃ 0.91, and dxd 1
|
f (x) x=1
≃ 0.96. Then the function f (x)
1
is
d 1
convex and dx f (x) is increasing function. Thus we get

d 1 d 1 G′ (x + 2)eG(x+2)
< lim = − lim .
dx f (x) x→∞ dx f (x) x→∞ (eG(x+2) − 1)2

Using the limit (20), we obtain

d 1
< 1, x > 0.
dx f (x)
Then q(x) is strictly decreasing function for all x > 0, where dq(x)
dx
= d 1
dx f (x)
− 1 < 0. Hence

lim q(x) < q(x) < lim+ q(x)

x→∞ x→0

and using the limit (19) and G(2) = 2 − ln 4, we have

4
1 < q(x) < . (22)
e2 −4
with best bounds.
1
In the proof of theorem (4), we proved that the function f (x) is convex. Also, the second
1
derivatives of the functions q(x) and f (x) have the same sign, then we get the following resuts:
Corollary 3.2. The function q(x) is strictly decreasing and convex for all x > 0.
( )
Corollary 3.3. For every x > 0 there exists a unique number µ ∈ 1, e24−4 such that G(x) =
( )
M (µ, x). Conversely for every µ ∈ 1, e24−4 there exists a unique number x > 0 such that
M (µ, x) = G(x).
( )
Proof. The function q(x) is strictly decreasing from (0, ∞) onto 1, e24−4 then the mapping
( )
q(x) : (0, ∞) → 1, e24−4 is bijective and the proof is easy consequence of this result.
( )
Corollary 3.4. For x > 2 and µ ∈ 1, e24−4 we have
( 2 )
−4)
1) the errors eµ (x) are uniformly bounded by ± ln 4(2e
3(3e2 −8)
.
( )
(x+2)[(e2 −4)x+4]
2) G(x) = M (µ, x) + O ln (x+1)[(e2 −4)x+e2 ]
.
( )
Proof. Analogues to inequality (17), we can deduce for all x > 2 and µ ∈ 1, e24−4 that
 ( )
 4 

0 ≤ |G(x) − M (µ, x)| ≤ M (1, x) − M , x 
e −4
2

which is equivalent to
   ( )
 (x + 2)[(e2 − 4)x + 4]   4(2e2
− 4) 
0 ≤ |eµ (x)| ≤ ln  ≤ ln .
(x + 1)[(e2 − 4)x + e2 ]   3(3e2 − 8) 

1173 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

4 Comparing approximations
Firstly, we will prove the following one side inequality the function G(x) which proves a special
case of a conjecture posed in [9] and proved in [11] about the best bounds of the Bateman’s
function but with different proof.
Lemma 4.1. For all x > 0, we have
1 1 1
G(x) − > 2 − 4. (23)
x 2x 4x
Proof. Consider the function
1 1 1
K(x) = G(x) − − 2 + 4, x > 0.
x 2x 4x
Using the integral representation (3) of G(x) and the formula
∫ ∞
1 1
= tr−1 e−xt dt, r∈N
xr (r − 1)! 0
we get ∫ ∞
e−xt
K(x) = φ(t) dt,
0 1 + et
where
1 1
φ(t) = et − 1 − t(1 + et ) + t3 (1 + et ).
2 24
But
∑
∞
tk 1 ∑ tk+1
∞
1 ∑ tk+3
∞
φ(t) = − +
k=4
k! 2 k=3 k! 24 k=1 k!
∑∞
t(k+4) 1
= (1 + (k + 4)[(k + 3)(k + 2) − 12])
k=0
(k + 4)! 24
∑∞
t(k+5) 1
= (1 + k(k + 5)(k + 7)) > 0.
k=0
(k + 5)! 24

Hence φ(x) > 0 and then K(x) > 0.

As by-product of the the inequalities (6) and (23), we obtain the following double inequality.
Corollary 4.2. For all x > 1, we have
(2x + 1)(x − 1)(x2 + 1) 2 1 1 2x2 − x + 1
0< < 2G(x) − − − < . (24)
2x4 (x + 1) x(x + 1) x 2x2 2x2 (x + 1)
Now, we will prove the following auxiliary results:
Lemma 4.3. For all x > x0 ≈ 2.5315129, we have
1 1
2
2G(x)− x(x+1) − x1 − 12
−x> 2x2 −x+1
− x > 1. (25)
e 2x −1 e 2x2 (x+1) −1

10

1174 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

Proof. Using the inequality (24), we have

( )
2 1 1 x+2
2G(x) − − − 2 − ln < u(x)
x(x + 1) x 2x x+1

where ( )
2x2 − x + 1 x+2
u(x) = − ln , x > 0.
2x2 (x + 1) x+1
Then √ √
(x − 3+ 17
)(x − 3− 17
)
u′ (x) = 2
3 2
2
,
x (x + 1)
√
3+ 17
and the function u(x) has only one positive critical point at xm = 2
. Now,
√
10 7 + 17
u(xm ) = √ − ln √ ≈ −0.00113 < 0,
(3 + 17)2 5 + 17

lim u(x) = 0
x→∞

and
lim u(x) = ∞.
x→0−

Hence u(x) has only one positive root x0 ≈ 2.5315129 and

u(x) < 0, ∀x > x0 .

Then ( )
2 1 1 x+2
2G(x) − − − 2 < ln , ∀x > x0 .
x(x + 1) x 2x x+1

Lemma 4.4. For all x > x1 ≈ 2.6925094, we have

1 4
−x< . (26)
e
2
2G(x)− x(x+1) − x1 − 1
2x2 −1 e2 −4

Proof. Using the inequality (24), we have

( )
2 1 1 e2 + (e2 − 4)x)
2G(x) − − − 2 − ln > v(x),
x(x + 1) x 2x 4 + (e2 − 4)x

where ( )
(2x + 1)(x − 1)(x2 + 1) e2 + (e2 − 4)x)
v(x) = − ln , x > 1.
2x4 (x + 1) 4 + (e2 − 4)x
Hence
L(x)
v ′ (x) = ,
S(x)

11

1175 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

where

L(x) = 8e2 + (−32 + 16e2 + 2e4 )x + (−32 − 12e2 + 6e4 )x2 + (48 − 36e2 + 5e4 )x3 + (32 − 4e2 )x4

+(−16 − 4e2 + e4 )x5 + (64 − 24e2 + 2e4 )x6

and
S(x) = x5 (x + 1)2 (4e2 + (e4 − 16)x + (16 − 8e2 + e4 )x2 > 0, x > 0.
The function L′′ (x) is a polynomial of fourth degree has one positive root at xI ≈ 2.31866 with
L′′ (3) < 0, then L(x) is concave function on (xI , ∞). Also, L(xI ) > 0 and limx→∞ L(x) = −∞.
Hence, the function L(x) has only one root on (xI , ∞) at x3 ≈ 4.0635204, where L(4.063) > 0
and L(4.064) < 0. Then L(x) > 0 on [xI , x3 ) and L(x) < 0 for all x > x3 . Hence v(x) is
increasing on (xI , x3 ) and decreasing function on (x3 , ∞) and it has a maximum point at x3 .
But v(2.69) < 0 and v(2.7) > 0 and then v(x) has a root x1 ≈ 2.6925094 ∈ (xI , x3 ). Also,
limx→∞ v(x) = 0, then we have
v(x) > 0, x > x1
and hence
( )
2 1 1 e2 + (e2 − 4)x)
2G(x) − − − 2 − ln > 0, x > x1 .
x(x + 1) x 2x 4 + (e2 − 4)x

[ ]
Theorem 5. For a fixed x > x1 , consider Ix be the nonempty open interval of 1, e24−4 defined
by ( )
1 1
Ix = − x, 2G(x)− 2 − 1 − 1 −x .
e− x(x+1) + x + 2x2 − 1
2 1 1
e x(x+1) x 2x2 − 1

For any µ ∈ Ix , we have  ( )

 1 1 

|eµ (x)| < G(x) − + .
x 2x2 
Proof. Using the inequalities (25) and (26), we obtain
[ ]
4
Ix ⊂ 1, 2 .
e −4
For any positive real number µ,
1 1 1
− x(x+1)
2
+ x1 + 12
− x < µ iff − M (µ, x) > − − 2
e 2x −1 x 2x

and hence
1 1 1
− x(x+1)
2
+ x1 + 12
− x < µ iff G(x) − M (µ, x) > G(x) − − 2. (27)
e 2x −1 x 2x

Also,
( )
1 2 1 1 1
− x > µ iff 2G(x) − − − < ln 1 +
e2G(x)− x(x+1) − x − 2x2
2 1 1
−1 x(x + 1) x 2x2 x+µ

12

1176 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

and hence
1 1 1
2
− x1 − 1 − x > µ iff G(x) − M (µ, x) < −G(x) + + 2. (28)
e2G(x)− x(x+1)
2x2 −1 x 2x

From the inequalities (27) and (28) we have

1 1 1 1
G(x) − − 2 < G(x) − M (µ, x) < −G(x) + + 2 , ∀µ ∈ Ix .
x 2x x 2x
Thus  ( )
 1 1 

|G(x) − M (µ, x)| < G(x) − + , ∀µ ∈ Ix . (29)
x 2x2 

References
[1] M. Abramowitz, I. A. Stegun (Eds.), Handbook of Mathematical Functions with Formulas,
Graphs and Mathematical Tables, Dover, New York, 1965.

[2] H. Alzer, On some inequalities for the gamma and psi function, Math. Comput., 66, 217,
373-389, 1997.

[3] G. E. Andrews, R.Askey and R.Roy, Special Functions, Cambridge Univ. Press, 1999.

[4] N. Batir, Inequalities for the gamma function, Arch. Math (Basel), 91, 554-563, 2008.

[5] N. Batir, An approximation formula for n!, Proyecciones Journal of Mathematics Vol. 32,
No 2, 173-181, 2013.

[6] A. Erdélyi et al., Higher Transcendental Functions Vol. I-III, California Institute of Tech-
nology - Bateman Manuscript Project, 1953-1955 McGraw-Hill Inc., reprinted by Krieger
Inc. 1981.

[7] B.-N. Guo and F. Qi, A class of completely monotonic functions involving divided differences
of the psi and tri-gamma functions and some applications, J. Korean Math. Soc. 48, 655-667,
2011.

[8] B.-N. Guo and F. Qi, Sharp inequalities for the psi function and harmonic numbers,
Analysis–International mathematical journal of analysis and its applications 34 , no. 2,
201-208, 2014.

[9] M. Mahmoud and R. P. Agarwal, Bounds for Bateman’s G−function and its applications,
Georgian Mathematical Journal, To appear 2016.

[10] M. Mahmoud, Some properties of a function related to a sequence originating from computa-
tion of the probability of intersecting between a plane couple and a convex body, Submitted
for publication.

13

1177 Mansour Mahmoud et al 1165-1178

 J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 23, NO.6, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC

[11] M. Mahmoud and H. Almuashi, On some inequalities of the Bateman’s G−function, J.

Comput. Anal. Appl., To appear 2017.

[12] M. Mansour, On quicker convergence towards Euler’s constant, J. Comput. Anal. Appl. 17,
No. 4, 632-638, 2014.

[13] C. Mortici, A sharp inequality involving the psi function, Acta Universitatis Apulensis,
41-45, 2010.

[14] C. Mortici, Estimating gamma function in terms of digamma function, Math. Comput.
Model., 52, no. 5-6, 942-946, 2010.

[15] C. Mortici, New approximation formulas for evaluating the ratio of gamma functions. Math.
Comp. Modell. 52(1-2), 425-433, 2010.

[16] C. Mortici, Accurate estimates of the gamma function involving the psi function, Numer.
Funct. Anal. Optim., 32, no. 4, 469-476, 2011.

[17] I. Muqattash and M. Yahdi, Infinite family of approximations of the Digamma function,
Mathematical and Computer Modelling 43, 13291336, 2006.

[18] K. Oldham, J. Myland and J. Spanier, An Atlas of Functions, 2nd edition. Springer, 2008.

[19] F. Qi, R.-Q. Cui, C.-P. Chen and B.-N. Guo, Some completely monotonic functions in-
volving polygamma functions and an application, Journal of Mathematical Analysis and
Applications 310, no. 1, 303-308, 2005.

[20] F. Qi and B.-N. Guo, Completely monotonic functions involving divided differences of the di-
and tri-gamma functions and some applications, Commun. Pure Appl. Anal. 8, 1975-1989,
2009.

[21] F. Qi and B.-N. Guo, Necessary and sufficient conditions for functions involving the tri- and
tetra-gamma functions to be completely monotonic, Adv. in Appl. Math. 44, 71-83, 2010.

[22] F. Qi and Q.-M. Luo, Bounds for the ratio of two gamma functions: from Wendel’s asymp-
totic relation to Elezović-Giordano-Pečarić’s theorem, Journal of Inequalities and Applica-
tions 2013, 2013:542, 20 pages.

[23] F. Qi , Bounds for the ratio of two gamma functions, J. Inequal. Appl. 2010 (2010), Article
ID 493058, 84 pages.

[24] S.-L. Qiu and M. Vuorinen, Some properties of the gamma and psi functions with applica-
tions, Math. Comp., 74, no. 250, 723-742, 2004.

14

1178 Mansour Mahmoud et al 1165-1178