Directory UMM :Journals:Journal_of_mathematics:OTHER:
1
2
3
47
6
Journal of Integer Sequences, Vol. 13 (2010),
Article 10.4.1
23 11
Asymptotic Expansions of Certain Sums
Involving Powers of Binomial Coefficients
Chun-Xue Zhao1 and Feng-Zhen Zhao1
School of Mathematical Sciences
Dalian University of Technology
Dalian 116024
China
[email protected]
[email protected]
Abstract
In this paper, we investigate the computation of sums involving binomial coefficients. We give asymptotic expansions of certain sums concerning powers of binomial
coefficients by using the multi-Laplace asymptotic integral theorem.
1
Introduction
Sums related to binomial coefficients are involved in many subjects such as combinatorial
analysis, graph theory, number theory, statistics and probability. For example, there is an
application of sums involving inverses of binomial coefficients (see [4]). For the computation
of sums concerning inverses of binomial coefficients, we may see for instance [10, 12, 13,
14, 15, 16]. Hence, it is important to discuss the computation of sums concerning binomial
coefficients. In this paper, we study the computation of certain sums involving powers of
binomial coefficients.
For convenience, we first give some notation and definitions involved in this paper. The
n
is defined by
binomial coefficient m
n!
, n ≥ m;
n
= m!(n − m)!
m
0,
n < m.
1
This work was supported by the Science Research Foundation of Dalian University of Technology (2008).
1
where n and m are nonnegative integers.
Throughout this paper, Hn (n ≥ 1) stands for the harmonic number, and the definition
of Hn (n ≥ 0) is
H0 = 0,
Hn =
n
X
1
k=1
k
,
n = 1, 2, . . . .
The generating function of Hn is
∞
X
Hn tn = −
n=1
ln(1 − t)
.
1−t
The harmonic numbers have been generalized by several authors (see [1, 2, 5, 6, 7, 8, 9]):
Hn[0]
1
= ,
n
Hn[r]
=
n
X
[r−1]
Hk
(n, r ≥ 1),
k=1
Hn,0 = 1, Hn,r =
X
1≤n1 0) is absolutely integrable in D;
(iii) The function f (x) reaches the effective maximum at the interior point ε = (ε1 , ε2 , . . . , εn )
in D, i.e., fk (ε) = 0, Hk [−f (ε)] > 0 (k = 1, 2, . . . , n), where, Hk [−f (x)] = det(−fij (x))1≤i,j≤k ;
(iv) The function φ(x) is continuous at point ε and φ(ε) 6= 0.
Then when λ → ∞,
ZZ
···
D
Z
n
φ(ε)[f (ε)]λ+ 2
φ(x)[f (x)] dx1 dx2 · · · dxn ∼ p
Hn [−f (ε)]
λ
2π
λ
n2
.
In this paper, we discuss the computation of certain sums related to powers of binomial
coefficients. In Section 2, we study the asymptotic expansions of sums such as
∞
X
n=0
1
(2n + 2k + 1)2
,
2n+2k 2
n+k
and in Section 3, we investigate the asymptotic expansions of sums involving binomial coefficients and generalized harmonic numbers.
2
Asymptotic Expansions of a Class of Sums Involving
Binomial Coefficients
In this section, we give the asymptotic expansions of sums involving inverses of binomial
coefficients.
Theorem 2. Let t, p, j and k be integers with p > 0, j ≥ 0 and k ≥ 0, we have the
asymptotic expansions as follows:
(i) when k is fixed,
Pp
n=0
(np )
2 ∼
(2n+2k+1)2 (2n+2k
n+k )
3
17p+1 π
42k+2p+1 p
(p → ∞),
(5)
(ii) when p is fixed,
(np )
Pp
n=0 (2n+2k+1)2 2n+2k 2
( n+k )
(iii)
∼
17p π
42k+2p+1 k
(k → ∞),
(6)
∼
π
42k−1 15k
(k → ∞),
(7)
P∞
1
n=0 (2n+2k+1)2 2n+2k 2
( n+k )
(iv) when k is fixed and k ≥ 1,
(n+k−1
)
n
P∞
n=0 (2n+2j+1)2 2n+2j 2
( n+j )
(v)
(n+k
n )
P∞
n=0 (2n+1)2 2n 2
(n)
(vi)
n(n+k
n )
P∞
n=0 (2n+1)2 2n 2
(n)
∼
∼
∼
π
42j−2k+1 15k j
42k+1 π
15k (k+1)
42k+1 π(k+1)
15k+1 (k+2)
(j → ∞),
(k → ∞),
(k → ∞).
(8)
(9)
(10)
Proof. In the proof, we will use the following formulas:
n
X
n k
x = (1 + x)n ,
k
k=1
∞
X
(11)
xn =
1
,
1−x
(12)
xn =
1
,
(1 − x)k
(13)
xn =
1
,
(1 − x)k+1
(14)
(k + 1)x
n+k n
.
n
x =
k+2
(1
−
x)
n
n=0
(15)
n=0
∞
X
n+k−1
n
n=0
∞
X
n+k
n=0
∞
X
n
4
It follows from (4) and (11)–(15) that
p
p
X
n
2
(2n + 2k + 1)2 2n+2k
n+k
Z 1
X p Z 1
n+k
n+k
y n+k (1 − y)n+k dy
=
x (1 − x)
dx
n 0
0
n=0
Z
Z
p
1
1
X
p
[xy(1 − x)(1 − y)]n+k dxdy
=
n
0
0
n=0
Z 1Z 1
p
X
p
k
(xy(1 − x)(1 − y))
=
(xy(1 − x)(1 − y))n dxdy
n
0
0
n=0
Z 1Z 1
=
(xy(1 − x)(1 − y))k (1 + xy(1 − x)(1 − y))p dxdy,
n=0
p
0
0
∞
X
=
=
1
n=0 (2n + 2k
∞ Z 1Z 1
X
n=0 0
1Z 1
Z
0
∞
X
0
(xy(1 − x)(1 − y))k
dxdy,
1 − xy(1 − x)(1 − y)
n+k−1
n
(2n + 2j + 1)2 2n+2j
n+j
Z 1Z 1X
∞
n+k−1
n=0
0
0
=
Z
0
1Z
1
0
Z
0
2
(xy(1 − x)(1 − y))n+j dxdy
(xy(1 − x)(1 − y))j
dxdy,
(1 − xy(1 − x)(1 − y))k
n+k
n
2 2n 2
(2n
+
1)
n=0
n
Z 1Z 1X
∞
n+k
0
=
n
n=0
∞
X
=
2n+2k 2
n+k
[xy(1 − x)(1 − y)]n+k dxdy
0
=
+ 1)2
0
1Z
0
n=0
1
n
(xy(1 − x)(1 − y))n dxdy
1
dxdy,
(1 − xy(1 − x)(1 − y))k+1
5
∞
X
n
n+k
n
2
(2n + 1)2 2n
n
Z 1Z 1X
∞
n+k
=
(xy(1 − x)(1 − y))n dxdy
n
n
0 n=0
0
Z 1Z 1
(k + 1)x(1 − x)y(1 − y)
=
dxdy.
k+2
0 (1 − xy(1 − x)(1 − y))
0
n=0
(i) Let φ(x, y) = (xy(1 − x)(1 − y))k , f (x, y) = 1 + xy(1 − x)(1 − y). It is evident that
f (x, y) > 0 for (x, y) ∈ [0, 1] × [0, 1],
fx
fy
fxx
fyy
fxy
=
=
=
=
=
y − 2xy − y 2 + 2xy 2 ,
x − x2 − 2xy + 2x2 y,
2y 2 − 2y,
2x2 − 2x,
1 − 2x − 2y + 4xy.
When fx = fy = 0, we get x = 21 , y = 21 . Then we have
h
1 1 i
1 1
H1 − f ,
= −fxx ,
2 2
2 2
1
,
=
2
h
1 1 i
−fxy 12 , 21
−fxx 21 , 21
H2 − f ,
=
2 2
−fyy 12 , 21
−fyx 12 , 21
1
0
= 2 1
0 2
1
=
,
4
1 1
= 4−2k ,
φ ,
2 2
1 1
17
f ,
.
=
2 2
16
By using Lemma 1, we have
Z 1Z 1
(xy(1 − x)(1 − y))k (1 + xy(1 − x)(1 − y))p dxdy
0
0
4−2k (17/16)p+1 2π
×
(p → ∞)
∼
1/2
p
17p+1 π
(p → ∞).
= 2k+2p+1
4
p
6
(ii) Let φ(x, y) = (1 + xy(1 − x)(1 − y))p , f (x, y) = xy(1 − x)(1 − y). It is evident that
y − 2xy − y 2 + 2xy 2 ,
x − x2 − 2xy + 2x2 y,
2y 2 − 2y,
2x2 − 2x,
1 − 2x − 2y + 4xy,
f (x, y) > 0,
(x, y) ∈ [η, 1 − η] × [η, 1 − η],
fx
fy
fxx
fyy
fxy
=
=
=
=
=
1
0 0, j ≥ 0 and k ≥ 0, we have the
asymptotic expansions as follows:
(i) when k is fixed,
3
3
p
p
X
65p+ 2 π 2
n
3 ∼ 6k+6p+3 3 (p → ∞),
3 2n+2k
2
p2
n=0 (2n + 2k + 1)
n+k
8
p
A1,p
B1,p
A1,p /B1,p
10
0.0018
0.0096
0.1832
50
0.0122
0.0216
0.5640
100 3.0777
3.5835
0.8589
150 44.8969 49.5072
0.9069
200 715.9654 769.4432
0.9305
240 6828.1
7246.9
0.9422
Table 1: some values of A1,p and B1,p
(ii) when p is fixed,
p
X
n=0
p
n
3
(2n + 2k + 1)3
(iii)
∞
X
n=0
n=0
(2n + 2k + 1)3
(2n + 2j + 1)3
n=0
n+k
n
n=0
n+k
n
(2n + 1)3
(k → ∞),
3
63k 2 26k−3
3
2n+2j 3
n+j
2n 3
n
(2n + 1)3
n
π2
∼
π2
3
26j−6k+3 63k j 2
(j → ∞),
3
(vi)
∞
X
∼
2n+2k 3
n+k
n+k−1
n
∞
X
(k → ∞),
3
26k+6p+3 k 2
3
(v)
3
∼
1
(iv) when k is fixed and k ≥ 1,
∞
X
2n+2k 3
n+k
65p π 2
2n 3
n
∼
π 2 26k+3
1
3
63k− 2 (k + 1) 2
(k → ∞),
3
∼
π 2 26k+3 (k + 1)
1
3
63k+ 2 (k + 2) 2
(k → ∞).
Asymptotic Expansions of Certain Sums Involving
Binomial Coefficients and Generalized Harmonic Numbers
There are many identities between the harmonic number and the binomial coefficient. For
example (see [11]),
∞
X
k
Hn
=
(k > 1).
n+k
2
(k
−
1)
k
n=1
9
Now we give asymptotic expansions of certain sums involving the generalized harmonic
numbers and the binomial coefficients.
Theorem 4. Let r be a positive integer, when r → ∞, we have the following asymptotic
expansions:
(i)
P∞
n=1
(ii)
[r]
Hn
P∞
n=r+1 Hn,r
(iii)
1
2
(2n+1)2 (2n
n)
(−1)n
2
(2n+1)2 2n
n
( )
P∞
n=r+1 H(n, r)
(−1)n
2
(2n+1)2 2n
n
( )
∼
∼
∼
Proof. It follows from (1)–(4) that
∞
X
=
=
n=1
∞
X
0
=
=
0
=
0
1
.
(−1)n
0
1Z
1
0
H(n, r)
Z
r+2
(17)
− ln (1 − xy(1 − x)(1 − y))
dxdy,
(1 − xy(1 − x)(1 − y))r
0
Z
n=r+1
Z
ln 17
16
,
2
(2n + 1)2 2n
n
Z 1Z 1
Hn,r (−1)n
[xy(1 − x)(1 − y)]n dxdy
(−1)
r!
∞
X
(−1)r+1 4π
r+1
r+1
0
0
Hn,r
n=r+1
r
ln 17
16
(16)
2
(2n + 1)2 2n
n
Z 1Z 1
[r]
[xy(1 − x)(1 − y)]n dxdy
Hn
n=1
1Z 1
n=r+1
∞
X
(−1)r 4π
rr!
16
ln 15
,
1
Hn[r]
Z
∞
X
16r−1 4π
15r−1 r
1
∞
X
lnr (1 + xy(1 − x)(1 − y))
dxdy,
1 + xy(1 − x)(1 − y)
(−1)n
(2n + 1)2
2n 2
n
H(n, r)(−xy(1 − x)(1 − y))n dxdy
0 n=r+1
Z 1Z 1
r+1
=(−1)
0
0
0
lnr+1 (1 + xy(1 − x)(1 − y))
dxdy.
1 + xy(1 − x)(1 − y)
10
(18)
Now we only give the detailed proof of (17), the others can be proved by the same method.
Let
1
f (x, y) = ln (1 + xy(1 − x)(1 − y)), φ(x, y) =
.
1 + xy(1 − x)(1 − y)
When (x, y) ∈ [η, 1 − η] × [η, 1 − η] (0 < η < 21 ), f (x, y) > 0, fx , fy , fxx , fyy , fxy can be
computed easily. When fx = 0, fy = 0, we get x = 21 , y = 12 . Then we have
h
1 1 i
H1 − f ,
2 2
h
1 1 i
H2 − f ,
2 2
1 1
φ ,
2 2
1 1
f ,
2 2
By using Lemma 1, we obtain
Z 1−η Z
1−η
η
η
8
,
17
2
8
=
,
17
16
,
=
17
17
= ln .
16
=
lnr (1 + xy(1 − x)(1 − y))
dxdy
1 + xy(1 − x)(1 − y)
17 r+1
(16/17)(ln 16
)
2π
×
(r → ∞)
8/17
r
r+1
17
4π
ln
(r → ∞).
=
r
16
∼
We can easily derive that
Z 1Z 1 r
Z 1−η Z 1−η r
ln (1 + xy(1 − x)(1 − y))
ln (1 + xy(1 − x)(1 − y))
dxdy −
dxdy
1 + xy(1 − x)(1 − y)
1 + xy(1 − x)(1 − y)
0
0
η
η
2
3
47
6
Journal of Integer Sequences, Vol. 13 (2010),
Article 10.4.1
23 11
Asymptotic Expansions of Certain Sums
Involving Powers of Binomial Coefficients
Chun-Xue Zhao1 and Feng-Zhen Zhao1
School of Mathematical Sciences
Dalian University of Technology
Dalian 116024
China
[email protected]
[email protected]
Abstract
In this paper, we investigate the computation of sums involving binomial coefficients. We give asymptotic expansions of certain sums concerning powers of binomial
coefficients by using the multi-Laplace asymptotic integral theorem.
1
Introduction
Sums related to binomial coefficients are involved in many subjects such as combinatorial
analysis, graph theory, number theory, statistics and probability. For example, there is an
application of sums involving inverses of binomial coefficients (see [4]). For the computation
of sums concerning inverses of binomial coefficients, we may see for instance [10, 12, 13,
14, 15, 16]. Hence, it is important to discuss the computation of sums concerning binomial
coefficients. In this paper, we study the computation of certain sums involving powers of
binomial coefficients.
For convenience, we first give some notation and definitions involved in this paper. The
n
is defined by
binomial coefficient m
n!
, n ≥ m;
n
= m!(n − m)!
m
0,
n < m.
1
This work was supported by the Science Research Foundation of Dalian University of Technology (2008).
1
where n and m are nonnegative integers.
Throughout this paper, Hn (n ≥ 1) stands for the harmonic number, and the definition
of Hn (n ≥ 0) is
H0 = 0,
Hn =
n
X
1
k=1
k
,
n = 1, 2, . . . .
The generating function of Hn is
∞
X
Hn tn = −
n=1
ln(1 − t)
.
1−t
The harmonic numbers have been generalized by several authors (see [1, 2, 5, 6, 7, 8, 9]):
Hn[0]
1
= ,
n
Hn[r]
=
n
X
[r−1]
Hk
(n, r ≥ 1),
k=1
Hn,0 = 1, Hn,r =
X
1≤n1 0) is absolutely integrable in D;
(iii) The function f (x) reaches the effective maximum at the interior point ε = (ε1 , ε2 , . . . , εn )
in D, i.e., fk (ε) = 0, Hk [−f (ε)] > 0 (k = 1, 2, . . . , n), where, Hk [−f (x)] = det(−fij (x))1≤i,j≤k ;
(iv) The function φ(x) is continuous at point ε and φ(ε) 6= 0.
Then when λ → ∞,
ZZ
···
D
Z
n
φ(ε)[f (ε)]λ+ 2
φ(x)[f (x)] dx1 dx2 · · · dxn ∼ p
Hn [−f (ε)]
λ
2π
λ
n2
.
In this paper, we discuss the computation of certain sums related to powers of binomial
coefficients. In Section 2, we study the asymptotic expansions of sums such as
∞
X
n=0
1
(2n + 2k + 1)2
,
2n+2k 2
n+k
and in Section 3, we investigate the asymptotic expansions of sums involving binomial coefficients and generalized harmonic numbers.
2
Asymptotic Expansions of a Class of Sums Involving
Binomial Coefficients
In this section, we give the asymptotic expansions of sums involving inverses of binomial
coefficients.
Theorem 2. Let t, p, j and k be integers with p > 0, j ≥ 0 and k ≥ 0, we have the
asymptotic expansions as follows:
(i) when k is fixed,
Pp
n=0
(np )
2 ∼
(2n+2k+1)2 (2n+2k
n+k )
3
17p+1 π
42k+2p+1 p
(p → ∞),
(5)
(ii) when p is fixed,
(np )
Pp
n=0 (2n+2k+1)2 2n+2k 2
( n+k )
(iii)
∼
17p π
42k+2p+1 k
(k → ∞),
(6)
∼
π
42k−1 15k
(k → ∞),
(7)
P∞
1
n=0 (2n+2k+1)2 2n+2k 2
( n+k )
(iv) when k is fixed and k ≥ 1,
(n+k−1
)
n
P∞
n=0 (2n+2j+1)2 2n+2j 2
( n+j )
(v)
(n+k
n )
P∞
n=0 (2n+1)2 2n 2
(n)
(vi)
n(n+k
n )
P∞
n=0 (2n+1)2 2n 2
(n)
∼
∼
∼
π
42j−2k+1 15k j
42k+1 π
15k (k+1)
42k+1 π(k+1)
15k+1 (k+2)
(j → ∞),
(k → ∞),
(k → ∞).
(8)
(9)
(10)
Proof. In the proof, we will use the following formulas:
n
X
n k
x = (1 + x)n ,
k
k=1
∞
X
(11)
xn =
1
,
1−x
(12)
xn =
1
,
(1 − x)k
(13)
xn =
1
,
(1 − x)k+1
(14)
(k + 1)x
n+k n
.
n
x =
k+2
(1
−
x)
n
n=0
(15)
n=0
∞
X
n+k−1
n
n=0
∞
X
n+k
n=0
∞
X
n
4
It follows from (4) and (11)–(15) that
p
p
X
n
2
(2n + 2k + 1)2 2n+2k
n+k
Z 1
X p Z 1
n+k
n+k
y n+k (1 − y)n+k dy
=
x (1 − x)
dx
n 0
0
n=0
Z
Z
p
1
1
X
p
[xy(1 − x)(1 − y)]n+k dxdy
=
n
0
0
n=0
Z 1Z 1
p
X
p
k
(xy(1 − x)(1 − y))
=
(xy(1 − x)(1 − y))n dxdy
n
0
0
n=0
Z 1Z 1
=
(xy(1 − x)(1 − y))k (1 + xy(1 − x)(1 − y))p dxdy,
n=0
p
0
0
∞
X
=
=
1
n=0 (2n + 2k
∞ Z 1Z 1
X
n=0 0
1Z 1
Z
0
∞
X
0
(xy(1 − x)(1 − y))k
dxdy,
1 − xy(1 − x)(1 − y)
n+k−1
n
(2n + 2j + 1)2 2n+2j
n+j
Z 1Z 1X
∞
n+k−1
n=0
0
0
=
Z
0
1Z
1
0
Z
0
2
(xy(1 − x)(1 − y))n+j dxdy
(xy(1 − x)(1 − y))j
dxdy,
(1 − xy(1 − x)(1 − y))k
n+k
n
2 2n 2
(2n
+
1)
n=0
n
Z 1Z 1X
∞
n+k
0
=
n
n=0
∞
X
=
2n+2k 2
n+k
[xy(1 − x)(1 − y)]n+k dxdy
0
=
+ 1)2
0
1Z
0
n=0
1
n
(xy(1 − x)(1 − y))n dxdy
1
dxdy,
(1 − xy(1 − x)(1 − y))k+1
5
∞
X
n
n+k
n
2
(2n + 1)2 2n
n
Z 1Z 1X
∞
n+k
=
(xy(1 − x)(1 − y))n dxdy
n
n
0 n=0
0
Z 1Z 1
(k + 1)x(1 − x)y(1 − y)
=
dxdy.
k+2
0 (1 − xy(1 − x)(1 − y))
0
n=0
(i) Let φ(x, y) = (xy(1 − x)(1 − y))k , f (x, y) = 1 + xy(1 − x)(1 − y). It is evident that
f (x, y) > 0 for (x, y) ∈ [0, 1] × [0, 1],
fx
fy
fxx
fyy
fxy
=
=
=
=
=
y − 2xy − y 2 + 2xy 2 ,
x − x2 − 2xy + 2x2 y,
2y 2 − 2y,
2x2 − 2x,
1 − 2x − 2y + 4xy.
When fx = fy = 0, we get x = 21 , y = 21 . Then we have
h
1 1 i
1 1
H1 − f ,
= −fxx ,
2 2
2 2
1
,
=
2
h
1 1 i
−fxy 12 , 21
−fxx 21 , 21
H2 − f ,
=
2 2
−fyy 12 , 21
−fyx 12 , 21
1
0
= 2 1
0 2
1
=
,
4
1 1
= 4−2k ,
φ ,
2 2
1 1
17
f ,
.
=
2 2
16
By using Lemma 1, we have
Z 1Z 1
(xy(1 − x)(1 − y))k (1 + xy(1 − x)(1 − y))p dxdy
0
0
4−2k (17/16)p+1 2π
×
(p → ∞)
∼
1/2
p
17p+1 π
(p → ∞).
= 2k+2p+1
4
p
6
(ii) Let φ(x, y) = (1 + xy(1 − x)(1 − y))p , f (x, y) = xy(1 − x)(1 − y). It is evident that
y − 2xy − y 2 + 2xy 2 ,
x − x2 − 2xy + 2x2 y,
2y 2 − 2y,
2x2 − 2x,
1 − 2x − 2y + 4xy,
f (x, y) > 0,
(x, y) ∈ [η, 1 − η] × [η, 1 − η],
fx
fy
fxx
fyy
fxy
=
=
=
=
=
1
0 0, j ≥ 0 and k ≥ 0, we have the
asymptotic expansions as follows:
(i) when k is fixed,
3
3
p
p
X
65p+ 2 π 2
n
3 ∼ 6k+6p+3 3 (p → ∞),
3 2n+2k
2
p2
n=0 (2n + 2k + 1)
n+k
8
p
A1,p
B1,p
A1,p /B1,p
10
0.0018
0.0096
0.1832
50
0.0122
0.0216
0.5640
100 3.0777
3.5835
0.8589
150 44.8969 49.5072
0.9069
200 715.9654 769.4432
0.9305
240 6828.1
7246.9
0.9422
Table 1: some values of A1,p and B1,p
(ii) when p is fixed,
p
X
n=0
p
n
3
(2n + 2k + 1)3
(iii)
∞
X
n=0
n=0
(2n + 2k + 1)3
(2n + 2j + 1)3
n=0
n+k
n
n=0
n+k
n
(2n + 1)3
(k → ∞),
3
63k 2 26k−3
3
2n+2j 3
n+j
2n 3
n
(2n + 1)3
n
π2
∼
π2
3
26j−6k+3 63k j 2
(j → ∞),
3
(vi)
∞
X
∼
2n+2k 3
n+k
n+k−1
n
∞
X
(k → ∞),
3
26k+6p+3 k 2
3
(v)
3
∼
1
(iv) when k is fixed and k ≥ 1,
∞
X
2n+2k 3
n+k
65p π 2
2n 3
n
∼
π 2 26k+3
1
3
63k− 2 (k + 1) 2
(k → ∞),
3
∼
π 2 26k+3 (k + 1)
1
3
63k+ 2 (k + 2) 2
(k → ∞).
Asymptotic Expansions of Certain Sums Involving
Binomial Coefficients and Generalized Harmonic Numbers
There are many identities between the harmonic number and the binomial coefficient. For
example (see [11]),
∞
X
k
Hn
=
(k > 1).
n+k
2
(k
−
1)
k
n=1
9
Now we give asymptotic expansions of certain sums involving the generalized harmonic
numbers and the binomial coefficients.
Theorem 4. Let r be a positive integer, when r → ∞, we have the following asymptotic
expansions:
(i)
P∞
n=1
(ii)
[r]
Hn
P∞
n=r+1 Hn,r
(iii)
1
2
(2n+1)2 (2n
n)
(−1)n
2
(2n+1)2 2n
n
( )
P∞
n=r+1 H(n, r)
(−1)n
2
(2n+1)2 2n
n
( )
∼
∼
∼
Proof. It follows from (1)–(4) that
∞
X
=
=
n=1
∞
X
0
=
=
0
=
0
1
.
(−1)n
0
1Z
1
0
H(n, r)
Z
r+2
(17)
− ln (1 − xy(1 − x)(1 − y))
dxdy,
(1 − xy(1 − x)(1 − y))r
0
Z
n=r+1
Z
ln 17
16
,
2
(2n + 1)2 2n
n
Z 1Z 1
Hn,r (−1)n
[xy(1 − x)(1 − y)]n dxdy
(−1)
r!
∞
X
(−1)r+1 4π
r+1
r+1
0
0
Hn,r
n=r+1
r
ln 17
16
(16)
2
(2n + 1)2 2n
n
Z 1Z 1
[r]
[xy(1 − x)(1 − y)]n dxdy
Hn
n=1
1Z 1
n=r+1
∞
X
(−1)r 4π
rr!
16
ln 15
,
1
Hn[r]
Z
∞
X
16r−1 4π
15r−1 r
1
∞
X
lnr (1 + xy(1 − x)(1 − y))
dxdy,
1 + xy(1 − x)(1 − y)
(−1)n
(2n + 1)2
2n 2
n
H(n, r)(−xy(1 − x)(1 − y))n dxdy
0 n=r+1
Z 1Z 1
r+1
=(−1)
0
0
0
lnr+1 (1 + xy(1 − x)(1 − y))
dxdy.
1 + xy(1 − x)(1 − y)
10
(18)
Now we only give the detailed proof of (17), the others can be proved by the same method.
Let
1
f (x, y) = ln (1 + xy(1 − x)(1 − y)), φ(x, y) =
.
1 + xy(1 − x)(1 − y)
When (x, y) ∈ [η, 1 − η] × [η, 1 − η] (0 < η < 21 ), f (x, y) > 0, fx , fy , fxx , fyy , fxy can be
computed easily. When fx = 0, fy = 0, we get x = 21 , y = 12 . Then we have
h
1 1 i
H1 − f ,
2 2
h
1 1 i
H2 − f ,
2 2
1 1
φ ,
2 2
1 1
f ,
2 2
By using Lemma 1, we obtain
Z 1−η Z
1−η
η
η
8
,
17
2
8
=
,
17
16
,
=
17
17
= ln .
16
=
lnr (1 + xy(1 − x)(1 − y))
dxdy
1 + xy(1 − x)(1 − y)
17 r+1
(16/17)(ln 16
)
2π
×
(r → ∞)
8/17
r
r+1
17
4π
ln
(r → ∞).
=
r
16
∼
We can easily derive that
Z 1Z 1 r
Z 1−η Z 1−η r
ln (1 + xy(1 − x)(1 − y))
ln (1 + xy(1 − x)(1 − y))
dxdy −
dxdy
1 + xy(1 − x)(1 − y)
1 + xy(1 − x)(1 − y)
0
0
η
η