J ournal of

Classical
A nalysis
Volume 15, Number 1 (2019), 9–22 doi:10.7153/jca-2019-15-02

NEW INTERESTING EULER SUMS

A MRIK S INGH N IMBRAN AND A NTHONY S OFO ∗

Abstract. We present here some new and interesting Euler sums obtained by means of related
integrals and elementary approach. Wesupplement Euler’s  general recurrence formula with two
(m) (m)
general formulas of the form ∑ On 1
(2n−1) p +
1
(2n) p and ∑ (2n−1)Op (2n+1)
n
q , where On =
n1 n1
n
1
∑ (2 j−1)m
. Two formulas for ζ (5) are also derived.
j=1

1. Evolution of Euler sums

Euler, the most prolific and versatile mathematician, made notable contributions
to all branches of mathematics but much of his significant work involves infinite series,
especially his zeta function.

1.1. Euler sums

In response to a letter of 24th December, 1742 from Goldbach, Euler investigated
sums involving zeta function and harmonic numbers and published his results many
years later in [7] using a cumbersome notation:
      
1 1 1 1 1 1 1
= 1 + 1 + + 1 + + + ....
zm yn 2m 2n 3m 2n 3n
Classical Euler sum BW (p, q) is thus an infinite sum whose general term contains gen-
n
(p) 1
eralized harmonic number Hn = ∑ p in numerator and nq with q  2 in denom-
k=1 k
∞ (p)
Hn
inator. That is, BW (p, q) = q ∑
. The number p + q is the weight of BW (p, q) .
n=1 n
In the said paper, Euler employed three methods (which he called Prima Methodus,
Secunda Methodus and Tertia Methodus) to discover formulas representing these sums
in terms of zeta values. First, he multiplied the involved series to obtain the reflection
formula:      
  
1 1 1 1 1 1 1
m n
+ n m
= m n
+ m+n
z y z y z z z
Mathematics subject classification (2010): 11M06, 11M32, 11Y60, 33B15, 40A25, 40B05.
Keywords and phrases: Euler sums, harmonic numbers, gamma function, psi function, zeta function,
polylogarithm function.
∗
Corresponding author.

c  , Zagreb 9
Paper JCA-15-02
10 A. S. N IMBRAN AND A. S OFO

which can be written as: BW (n, m) + BW (m, n) = ζ (m) ζ (n) + ζ (m + n)

∞ (m)
Hn 1 1
and straightway leads to the sum: ∑ m = ζ 2 (m) + ζ (2m). Euler gave a general
n=1 n 2 2
formula (without proof) expressing EU (m) = BW (1, m) in terms of zeta values. We
find in [7, §22] a list of formulas which, for m  2, can be written as:
∞
Hn m + 2 1 m−2
EU (m) = ∑ m = ζ (m + 1) − ∑ ζ (m − k) ζ (k + 1), m = 2, 3, 4, . . .. (1.1)
n=1 n 2 2 k=1

1.2. Post-Euler development

These recurrence relations hold [21]:

2 m−1
ζ (2m) = ∑ ζ (2k)ζ (2m − 2k), m = 2, 3, 4, . . . ,
2m + 1 k=1
m
2
(1 − 2−2m)ζ (2m) = ∑ ξ (2k − 1)ξ (2m + 1 − 2k), m = 1, 2, 3, . . . ,
2m − 1 k=1

where ξ (2m + 1) := ∑∞ n=1 (−1)

n−1 1
(2n−1)2m+1
.
Nielsen [11, pp.47–49] built on and supplemented Euler’s work by supplying proof
of the general formula using the method of partial fractions.
Georghiou and Philippou [9] established formula (1.1) and
∞ 2n
Hk 1
∑ k2n+1 = 2 ∑ (−1) j ζ ( j) ζ (2n + 2 − j), n  1. (1.2)
k=1 j=2

For odd weight p + q = 2r + 1 , Borweins [3] (see[8, Theorem 3.1]) established a

correct version of Euler’s formula relating to exponent  2 of the harmonic series:

∞ (p)     
Hn 1 p 2r p 2r
BW (p, q) = ∑ q
= 1 − (−1) − (−1) ζ (2r + 1)
n=1 n 2 p q
p/2  
1 − (−1) p 2r − 2k
+ ζ (p)ζ (q) + (−1) p ∑ ζ (2k)ζ (2r + 1 − 2k)
2 k=1 q−1
q/2  
2r − 2k
+ (−1) ∑p
ζ (2k)ζ (2r + 1 − 2k + 1).
k=1 p−1
(1.3)

The sum BW (p, q) also admits of representation in terms of zeta values when: p =
q, and (p, q) = (2, 4) or (4, 2). For alternating Euler sums, Sitaramachandra Rao [15]
gave the identity:
(p)
∞
(−1)n+1 Hn
S (p, q) := ∑ ,
n =1 nq
 E ULER SUMS 11

when p = 1 and for odd weight 1 + q as,

q
2 −1
2S (1, q) = (1 + q) η (1 + q) − ζ (1 + q) − 2 ∑ η (2 j) ζ (1 + q − 2 j). (1.4)
j=1

Flajolet and Salvy [8] also gave the integral

 1 2q
1 ln (x) ln (1 + x)
S (1, 1 + 2q) = dx.
(2q)! 0 x (1 + x)
In the case where p and q are both positive integers and p + q is an odd integer, they
also [8] gave the identity:
 
p+i−1
2S (p, q) = (1 − (−1) p ) ζ (p) η (q) + 2 (−1) p ∑ ζ (p + i) η (2k)
i+2k=q
p−1
 
q+ j−1
+ η (p + q) − 2 ∑ (−1) j η (q + j) η (2k) , (1.5)
j+2k=p
q − 1

where η (0) = 12 , η (1) = ln 2, ζ (1) = 0, and ζ (0) = − 12 in accordance with the ana-
lytic continuation of the Riemann zeta function. We define the alternating zeta function
(or Dirichlet eta function) η (z) as
∞
(−1)n+1

η (z) := ∑ nz = 1 − 21−z ζ (z) .

n =1

Sofo [17], further, developed the half integer Euler sums. For positive integers m, p and
odd weight m + p ,
⎛ ⎞
Hn
(m)
m   BW (r, m + p − r)
m+ p−1−r ⎝ ⎠
W (m, p) = ∑ 2p = (−1) p ∑ 2m−1
n m−r
n1 r=1 −S (r, m + p − r)
m  
1 m+ p−1−r
+ (−1) p+1 ∑ p−r ζ (r) ζ (m + p − r) (1.6)
r=2 2
m−r
 
p−1
(−1)k m+ p−1−k
+ (−1) p+1 ∑ p−k p−k
ζ (k) ζ (m + p − k).
k=2 2

Also, since
(m)
Hn
(m) Hn  
∑ np =2 p−1
∑ 2
np
1 − (−1)n+1 ,
n1 n1

we obtain the alternating Euler identity at half integer value,

(m)
(−1)n+1 H n
AW (m, p) = ∑ np
2
= W (m, p) − 21−pBW (m, p) . (1.7)
n1
12 A. S. N IMBRAN AND A. S OFO

It is interesting to note that for (1.6) or (1.7) we can evaluate the difference (or sum) of
two terms, in the following way. Let (m, p, r) be positive integers, with p  2, then
(m) (m)
Hn Hn
F (m, p, r) = W (m, p, r) − W (m, p, r + 2) = ∑ (n + r)p
2
−∑ 2
p
n1 n1 (n + r + 2)
= 2m ζ (m + p) − 2m η (m) , for r = 0.
For r > 0
 
m+ p−1− j ( j)
p Hr
p− j η (m)
−m
2 F (m, p, r) = (−1)m+1
∑ r m+p− j
−
(1 + r) p
j=1
   
m+ p−1− j m+ p−1−k
p ζ ( j) m ζ (k)
p− j m−k
+(−1)m ∑ +(−1) m
∑ .
j=2 rm+p− j k=2 rm+p−k
F (5, 4, 0) =32ζ (9) − 30ζ (5) ,
79 31
F (2, 4, 2) =ζ (4) + ζ (3) + ζ (2) − .
81 8

2. Euler sums involving odd harmonic numbers

n
1 1
Let On := ∑ 2k − 1 . Then On = H2n − 2 Hn .
k=1
One can compute the following sums with little effort:
∞
Hn
∑ (2n − 1)(2n + 1) = ln 2; (2.1)
n=1
∞
On 3
∑ (2n − 1)(2n + 1) = 8 ζ (2); (2.2)
n=1
∞
On 7
∑ n 2
= ζ (3).
4
(2.3)
n=1
∞
On 7
This sum noted in [15, p.14, eq(4.16)]: ∑ (−1)n+1 n2 = π G − ζ (3) when added to
4
n=1
(2.3) gives a nice double sum:
∞ 2n−1  
1 1 1 1 1 πG
∑ (2n − 1)2 ∑ 2k − 1 = 1 + 32 1+ +
3 5
+ ··· =
2
,
n=1 k=1
n+1
(−1)
where G = ∑∞ n=1 (2n−1)2 is Catalan’s constant. After some manipulation we can also
isolate the identity,
∞
H4n−2 21 πG
∑ (2n − 1)2 = 32 ζ (3) + 2
n=1
 E ULER SUMS 13

and the alternating double sum

   
∞
(−1)n+1 n (−1)k+1 1 1 1 1 1
∑ n ∑ 2k − 1 = 1 (1) − 2 1 − 3 + 3 1 − 3 + 5 − · · · = G.
n=1 k=1

Sofo derives this formula [16, Corollary 1] for real α = −1, −2, −3, . . .:
∞
Hn
2∑ =2ζ (m, α + 1)Hα + m ζ (m + 1, α + 1)
n=1 (n + α + 1)
m

m−2
− ∑ ζ (k + 1, α + 1)ζ (m − k, α + 1),
k=1

which leads to
∞
Hn (2m − 1) m (2m+1 − 1)
∑ m
= −
2m−1
ζ (m) ln 2 +
2m+1
ζ (m + 1)
n=1 (2n + 1)
m−2
(2k+1 − 1) (2m−k − 1)
− ∑ 2m+1
ζ (k + 1) ζ (m − k).
k=1

1
If we, let κ (m) = 12 (ζ (m) + η (m)) = ∑k1 (2k−1)m , and for even m only upon simpli-

fication one obtains

∞
Hn 1 m−2
∑ (2n + 1)m = m κ (m + 1) − κ (m) ln 2 − 2 ∑ κ (k + 1) κ (m − k) . (2.4)
n=1 k=1

By shifting index one gets sums with powers of (2n − 1) in the denominator, therefore
Hn
we are able to obtain an identity for ∑∞
n=1 (2n−1)m .

2.1. Formulas with On in numerator and odd factors in denominator

The following formulas, valid only for even m ∈ N, have been adapted from Jordan
[10, 1]:
m −1
∞
On 2m+1 −1 1 2
U(m) := ∑ m
= m+2
ζ (m+1)− m+1 ∑ (2m−2k+1 −1) ζ (2k) ζ (m−2k+1).
n=1 (2n) 2 2 k=1
(2.5)

Also, for even m only

∞
On 1
V (m) := ∑ m
= κ (m + 1) + κ (m) ln 2
n=1 (2n − 1) 2
m −1
1 2
−
2m+1 ∑ (22k − 1) ζ (2k) ζ (m − 2k + 1). (2.6)
k=1
14 A. S. N IMBRAN AND A. S OFO

From (2.6), we know that

∞
On ∞ H2n − 1 Hn
∑ (2n − 1)2m = ∑ (2n − 1)2 2m ,
n=1 n=1

hence we have the new identity

∞ ∞
H2n 1 Hn 22m+1 − 1 22m − 1
∑ (2n − 1)2m = 2 ∑ (2n − 1)2m + 22m+2
ζ (2m + 1) +
22m
ζ (2m) ln 2
n=1 n=1
m−1
1
−
22m+1 ∑ (22k − 1) ζ (2k) ζ (2m − 2k + 1).
k=1

We found a remarkable recurrence for odd powers with k ∈ N:

∞ ∞
On On
∑ (2n − 1)2k+1 + ∑ (2n)2k+1 (2.7)
n=1 n=1
k−1 2 j+1
1 2 −1
= κ (2k + 2) + κ (2k + 1)ln 2 − ∑ 2k+2
ζ (2 j + 1)ζ (2k − 2 j + 1).
2 j=1 2

Ramanujan [12, Ch.IX, p.104, 11.ii][2, p.257, (11.3)] recorded a wrong result:
∞ ∞ ∞
On π (−1)k π 1
∑ (2n)3 = 4 ∑ (4k + 1)3 − 3√3 ∑ (2k + 1)3 .
n=1 k=0 k=0

Sitaramachandrarao [15] notes:

∞ ∞ (2)
On 3 1 Hn
∑ (2n)3 = 16 ζ (4) − 4 ∑ (−1)n+1 (n + 1)2 .
n=1 n=1

The last term involves the constant

∞ n
1 1
A4 = ∑ (−1)n+1 (n + 1)2 ∑ k2 ≈ 0.1626546673974.
n=1 k=1

∞
On
So putting the value of ∑ (2n)3 in the first formula in the chain, we get:
n=1

∞
On 9 7 A4
∑ (2n − 1)3 = 32 ζ (4) + 8 ζ (3) ln 2 + 4
.
n=1

We find this value of the constant in [13]:

 
65 7 1 1
A4 = ζ (4) − ζ (3) ln 2 + ζ (2)(ln2)2 − (ln 2)4 − 4Li4 ,
16 2 6 2
 E ULER SUMS 15
  ∞
1 1
where Li4 = ∑ n 4 ≈ 0.517479061673899. We then have [15, p.3, corrected]:
2 n=1 2 n
∞  
On 53 7 1 1 1
∑ (2n)3 = − 64 ζ (4) + 8 ζ (3) ln 2 − 4 ζ (2)(ln 2) + 24 (ln 2) + Li4 2
2 4
n=1

and
∞  
On 83 1 1 1
LS (3) := ∑ (2n − 1)3 = 64 ζ (4) + 4 ζ (2)(ln 2)2 − 24 (ln 2)4 − Li4 2
, (2.8)
n=1

which through shift of index yields

∞  
On 23 1 1 1
∑ (2n + 1)3 = 64 ζ (4) + 4 ζ (2)(ln 2)2 − 24 (ln 2)4 − Li4 2
.
n=1

It is possible to generalize further the result (2.7) in the following way.

L EMMA 1. Let (m, p) be positive integers with p  2 and (m + p) an odd integer.

n
(m) 1
Put On := ∑ m . Then
k=1 (2k − 1)

(m) (m) 1 (m) 1 (m)

On = H2n − m
Hn = m H 1 + η (m) . (2.9)
2 2 n− 2
Hence
∞  
(m)1 1 1
X (m, p) := ∑ On (2n − 1) p
+
(2n) p
= κ (p) η (m) + m W (m, p)
2
n=1
 m+p 
2 −4 1
+ m+p+1
BW (m, p) − S (m, p) , (2.10)
2 2

where W (m, p) is given by (1.6), BW (m, p) is given by (1.3) and S (m, p) is given by
(1.5).

Proof. Consider, using (2.9)

⎛ (m) ⎞
  1
2m Hn− 1 + η (m)
(m) (m)
∞
(m) 1 1 ∞ H2n − 21m Hn
X (m, p) := ∑ On p
+ = ∑⎝ 2
+ ⎠
n=1 (2n − 1) (2n) p n=1 (2n − 1) p (2n) p
(m)
1 ∞ Hn− 21
∞ ∞ H (m) (m)
1 1 ∞ Hn
= η (m) ∑ p
+ ∑ + ∑ 2n
− ∑
n=1 (2n − 1) 2m n=1 (2n − 1) p n=1 (2n) p 2m n=1 (2n) p
⎛ (m) (m)
⎞
1 1 ∞ ⎝ H 2n (−1)n+1 H n
= m−1 κ (p) η (m) + m+1 ∑ + 2 ⎠
2 2 n=1 np np
16 A. S. N IMBRAN AND A. S OFO
 (m) (m)
 (m)
1 ∞ Hn (−1)n+1 Hn 1 ∞
Hn
+ ∑ − − ∑ np .
2 n=1 np np 2m+p n=1

Using (1.7) and simplifying,

1
1
X (m, p) = κ (p) η (m)+ 2W (m, p) − 21−pBW (m, p) + (BW (m, p) − S (m, p))
2m+1 2
1
− BW (m, p) ,
2m+p
therefore collecting terms, (2.10) follows. 

E XAMPLE 1.
 
∞
(2) 1 1 11 29 635
X (2, 5) = ∑ On 5
+ 5
= ζ (4) ζ (3)+ ζ (5) ζ (2)− ζ (7) ,
n =1 (2n − 1) (2n) 32 16 256

here we have used

(2)
Hn 21 107
∑ 2
n5
= ζ (4) ζ (3) −
16
ζ (5) ζ (2) +
64
ζ (7)
n1

and
(2)
(−1)n+1 H n 7 13 147
∑ n5
2
= ζ (4) ζ (3) − ζ (5) ζ (2) +
8 8 64
ζ (7) .
n1

 
∞
(3) 1 1 127 45 147
X (3, 4) = ∑ On
(2n−1)4
+
(2n)4
=
16
ζ (7)+ ζ (4) ζ (3)−
64 32
ζ (5) ζ (2) .
n =1

2.2. Sums with two odd factors in denominator

In this section we develop identities for Euler sums of the form
∞
On
Y (p, q) = ∑ (2n − 1) p(2n + 1)q , (2.11)
n=1

which in turn we extract further Euler like sums of the type

∞
H2n
∑ (4n2 − 1) p . (2.12)
n=1

L EMMA 2. For positive integers p, q and p  4, q  4 , we have

 
∞
On 3 (−1) p+1 p+q−2
Y (p, q) = ∑ = ζ (2)
p
n=1 (2n − 1) (2n + 1)
q 2 p+q+1 p−1
 E ULER SUMS 17
 
q
(−1) p+1 p + q − k − 1
+ ∑ p+q−k κ (k + 1)
k=2 2
q−k
q  
(−1) p p + q − k − 1
+ ∑ p+q−k V (k)
k=2 2
q−k
 
p
(−1) p− j p + q − j − 1
+ ∑ p+q− j V ( j) , (2.13)
j=2 2
p− j

where V (·) is given by (2.6).

Proof. Consider the partial fraction decomposition

 
1 (−1) p+1 p + q − 2 1
=
p
(2n − 1) (2n + 1) q 2 p+q−2 p−1 (2n − 1)(2n + 1)
p p− j  
(−1) p+q− j−1 1
+ ∑ p+q− j
j=2 2 p − j (2n − 1) j
 
q
(−1) p p + q − k − 1 1
+ ∑ p+q−k ,
k=2 2 q − k (2n + 1)k
now by summing over the integers n
∞
On
Y (p, q) = ∑ (2n − 1) p(2n + 1)q (2.14)
n=1
 
(−1) p+1 p+q−2 ∞ On
=
2 p+q−2 p−1 ∑
n=1 (2n − 1)(2n + 1)
p p− j  
(−1) p+q− j−1 ∞ On
+ ∑ p+q− j ∑
j=2 2
p − j n=1 (2n − 1)
j
 
q
(−1) p p + q − k − 1 ∞ On
+ ∑ p+q−k ∑ .
k=2 2
q−k n=1 (2n + 1)
k

Consider the third term in (2.14) and make a change in the summation index, so that
 
q
(−1) p p + q − k − 1 ∞ On
∑ 2 p+q−k q − k ∑ (2n + 1)k
k=2 n=1
q p   ∞
(−1) p+q−k−1 On
= ∑ p+q−k ∑ (2n − 1)k
k=2 2 q − k n=1
q p+1  
(−1) p+q−k−1
+ ∑ p+q−k κ (k + 1). (2.15)
k=2 2
q−k

Substituting (2.15) into (2.14), we have

   q  
(−1) p+1 p+q−2 3 (−1) p+1 p+q−k−1
Y (p, q) = p+q−2 ζ (2) + ∑ p+q−k κ (k+1)
2 p−1 8 k=2 2
q−k
18 A. S. N IMBRAN AND A. S OFO

   
(−1) p
q
p+q−k−1
p
(−1) p− j p + q − j − 1
+ ∑ p+q−k V (k) + ∑ p+q− j V ( j) ,
k=2 2
q−k j=2 2
p− j

where V (·) is given by (2.6), therefore the identity (2.13) follows. 

R EMARK 1. While the identity (2.13) is numerically correct for all integer values
p, q  1, we restrict its application to (p, q)  4 because we do not have closed form
identities of V (2m + 1), for m  2.

E XAMPLE 2.
∞
On 3 3 15 15
Y (2, 4) = ∑ (2n − 1)2(2n + 1)4 = 16 ζ (2)ln 2 − 32 ζ (2) + 64 ζ (4) ln 2 − 64 ζ (4)
n=1
1 7 31 1
+ LS (3) − ζ (3) − ζ (5) − ζ (2) ζ (3) ,
4 128 256 64
∞
On 3 3 15 7
Y (4, 2) = ∑ 4(2n + 1)2
= ζ (2)ln 2 − ζ (2) + ζ (4) ln 2 + ζ (3)
n=1 (2n − 1) 16 32 64 128
1 31 3
− LS (3) + ζ (5) − ζ (2) ζ (3) .
4 256 128
An important corollary of Lemma 2.13 is the case of q = p.

C OROLLARY 1. From (2.13), for the case q = p and p ∈ N\ {1} we have

 
∞
On 3 (−1) p+1 2p − 2
Y (p, p) = ∑ = ζ (2)
2
n=1 (4n − 1)
p 22p+1 p−1
 
p
(−1) p+1 2p − j − 1
+ ∑ 2p− j κ ( j + 1)
j=2 2
p− j
 
p
(−1) p 2p − j − 1  
+ ∑ 2p− j 1 + (−1) j V ( j) . (2.16)
j=2 2
p− j

R EMARK 2. From (2.16) we have,

∞
On ∞ H2n − 1 Hn
Y (p, p) = ∑ (4n2 − 1) p = ∑ (4n2 −21) p
n=1 n=1

from which we obtain the new identity

 
∞
H2n 1 ∞ Hn 3 (−1) p+1 2p − 2
∑ 2 p 2 ∑ (4n2 − 1) p
= +
22p+1 p−1
ζ (2)
n=1 (4n − 1) n=1
 
p
(−1) p+1 2p − j − 1
+ ∑ 2p− j κ ( j + 1)
j=2 2
p− j
 E ULER SUMS 19
  
p
(−1) p 2p − j − 1
+ ∑ 2p− j 1 + (−1) j V ( j) . (2.17)
j=2 2
p− j

We note that identities (2.16) and (2.17) hold for all integer p values bigger than one,
because we do not require identities of V (2m + 1) for m  2.

E XAMPLE 3. Some miscellaneous examples are highlighted.

∞
Hn 7 π2
∑ 2 3 = − ζ (3) − (5 − 6 ln2) + ln2, (2.18)
n=1 (4n − 1) 16 64
∞
Hn 31 π4 7
∑ (4n2 − 1)4 = 64 ζ (5) − 768 (2 ln 2 − 1) − ln2 − 256 ζ (3)(π 2 − 8)
n=1
π2
+ (11 − 10 ln2), (2.19)
128
∞
On π4 π2 5π 2
∑ (4n2 − 1)4 = 768 (ln 2 − 1) − 512 ζ (3) + 256 (2 ln 2 − 1),
n=1
∞
On π6 5π 4 5π 2 35π 2
∑ (4n2 − 1)5 = 30720 + 12288 (3 − 4 ln2) + 2048 ζ (3) − 2048 (2 ln 2 − 1),
n=1
(2.20)
∞  
On 189 315 63 189
∑ (4n2 − 1)6 = 1024 ζ (2) + 2048 ζ (4) + 2048 ζ (6) ln 2 − 4096 ζ (6)
n=1
 
189 63 3
− + ζ (3) + ζ (5) ζ (2)
2048 4096 4096
 
105 15
− + ζ (3) ζ (4) ,
1024 4096
∞
On 17 7 π2
∑ (2n − 1)3 (2n)3 = 8 ζ (4) − 8 ζ (3)(3 + ln2) + 24 (2 ln2 2 − 9 ln2 + 12)
n=1
 
ln4 2 1
− − 2Li4 , (2.21)
12 2
∞
H2n 15 7 3 1
∑ (4n2 − 1)3 = 128 ζ (4) − 32 ζ (3) − 32 ζ (2) + 2 ln 2,
n=1
∞
H2n 381 63 527 135 1
∑ (4n2 − 1)6 = 4096 ζ (7) − 2048 ζ (6) + 2048 ζ (5) + 4096 ζ (4) − 2 ln 2
n=1
7 195 3 15
− ζ (3) + ζ (2) − ζ (2)ζ (5) − ζ (4)ζ (3)
1024 1024 128 572
63
− ζ (2)ζ (3).
512
20 A. S. N IMBRAN AND A. S OFO

2.3. Euler sums involving powers of harmonic numbers

Euler did not treat sums that involve powers of harmonic numbers. H. F. Sandham,
an Irish mathematician1, introduced the first quadratic double sum in 1948 as a problem
in the American Mathematical Monthly [14].
∞  
Hn 2 17π 4
∑ n = 360 . (2.22)
n=1

Apparently, it went unnoticed. It was recorded by Castellanos in his 1988 survey article
[5, I, p.86], rightly attributed to H. F. Sandham but with a wrong entry in the bibliogra-
phy.
De Doelder [6] evaluated this associated sum in 1991 without any reference to
Sandham’s sum:
∞  
Hn 2 11π 4
∑ = . (2.23)
n=1 n + 1 360
For this, he used the psi function.
The comparison of the terms in the two series in the following expression makes
it clear that
∞ ∞ ∞ ∞ H2 − H2
H2 H2 H2
∑ n(n +n 1) = ∑ nn − ∑ n +n 1 = ∑ n n n−1 .
n=1 n=1 n=1 n=1

Now Hn2 − Hn−1

2
= (Hn − Hn−1)(Hn + Hn−1) = 1n (2Hn − 1n ) = 2 Hnn − n12 . Hence, we get
∞ ∞ ∞
Hn2 Hn 1
∑ n(n + 1) ∑ n2 ∑ n3 = 3ζ (3).
= 2 − (2.24)
n=1 n=1 n=1

We also have:
∞ 2
Hn+1
∑ = ζ (2) + 3.
n=1 n(n + 1)
We can deduce De Doelder’s sum from that of Sandham as follows. On comparing the
terms of the two series, we notice that
∞    2
Hn 2 ∞ Hn ∞ H2 − H2
∑ n − ∑ (n + 1) = ∑ n n2 n−1 .
n=1 n=1 n=1

Therefore,
∞  2 ∞  2 ∞ ∞
Hn Hn Hn 1 3 π4
∑ n
−∑
(n + 1)
=2∑ 3
− ∑ 4 = ζ (4) = .
n=1 n=1 n=1 n n=1 n 2 60
1 Henry Francis Sandham (1917–1963) studied mathematics at Trinity College, Dublin (Ireland) and
Queen’s University Belfast, received a Ph.D. in 1958 on his thesis Products of the Hypergeometric Func-
tions, and taught at Trinity College. In September 1952, he joined the School of Theoretical Physics as a
lecturer under the Nobel laureate Erwin Schrödinger at Dublin Institute for Advanced Studies which he left
in 1956 to work with English Electric Labs (Staffordshire) until his death in April 1963. In addition to half a
dozen papers and problems published in various journals, he presented a note on the Perimeter of an Ellipse
at the International Congress of Mathematicians, Amsterdam, September 2–9, 1954.
 E ULER SUMS 21

Sandham’s sum was conjectured in April 1993 by Enrico Au-Yeung, a student of J.

Borwein at the University of Waterloo, on the basis of his computations and was estab-
lished (by means of generating functions and Parseval’s identity for Fourier series) by
Borweins [4], who referred to De Doelder’s paper but not to Sandham. Other related
identities for Euler sums are available in [18], [19] and [20].

2.4. Few miscellaneous sums

De Doelder’s formulas [6, (15) and (22)] are worthy of note:

ψ ( 12 ± k) − ψ ( 12 ) 7
∞ ∞ (ψ ( 1 ± k) − ψ ( 1 ))2
π4
∑ k 2
= ζ (3), ∑
2
2
k 2
2
=
16
.
k=1 k=1
    
1 1 1
Now On = ψ n+ −ψ . Thus we get [4]:
2 2 2
∞  
∞
On 7 On 2 π 4
∑ 2 = 4 ζ (3), ∑ n = 32 .
n=1 n n=1

The first occurs in Ramanujan’s Manuscript2 [12, Ch. IX, p.104, Entry12, Ex.iii]. We
have this associated sum:
∞
On 7 3
∑ (2n − 1)2 = 16 ζ (3) + 4 ln 2ζ (2).
n=1

We could obtain:
∞
On 31 7
∑ n 4
= ζ (5) − ζ (3)ζ (2),
4 2
n=1
as well as
∞ ∞
H2n−1 155 3 H2n 37 1
∑ (2n − 1)4 = 64
ζ (5) − ζ (3)ζ (2),
4 ∑ (2n)4 = 64 ζ (5) − 4 ζ (3)ζ (2)
n=1 n=1

and
∞ ∞
H2n−1 35 1 H2n 93 3
∑ (2n)4 = 64 ζ (5) − 4 ζ (3)ζ (2), ∑ (2n + 1)4 = 64 ζ (5) − 4 ζ (3)ζ (2).
n=1 n=1

The previous four results yield two nice formulas:

∞
(1 + (−1)n2)Hn 11
∑ n 4
= − ζ (5);
16
(2.25)
n=1
∞
(1 − (−1)n2)Hn 3
∑ (n + 1)4
= ζ (5).
16
(2.26)
n=1

Acknowledgement. The authors are thankful to the referee for carefully considered
suggestions leading to a better presented paper.
22 A. S. N IMBRAN AND A. S OFO

REFERENCES

1
[1] G. BASTIEN , Elementary methods for evaluating Jordan’s sums ∑n1 1 + 13 + ··· + 2n−1 1
n2a
and

∑n1 1 + 13 + ··· + 2n−11 1

(2n−1)2a
and analogous Euler’s type sums and for setting a σ -sum theorem,
2013, http://arxiv.org/abs/1301.7662v1.
[2] B. C. B ERNDT, Ramanujan’s Notebooks Part 1, Springer-Verlag, New York 1985.
[3] D. B ORWEIN , J. M. B ORWEIN AND R. G IRGENSOHN, Explicit Evaluation of Euler Sums, Proceed-
ings of the Edinburgh Mathematical Society 38 (1995), 277–294.
[4] D. B ORWEIN AND J. M. B ORWEIN , On an intriguing integral and some series related to ζ (4) , Proc.
Amer. Math. Soc. 123 (1995), 1191–1198.
[5] D. C ASTELLANOS , The Ubiquitous π , Math. Mag. Part I, 61(2) 1988, 67–98; Part II, 61(3) 1988,
148–163.
[6] P. J. D E D OELDER, On some series containing ψ (x) − ψ (y) and (ψ (x) − ψ (y))2 for certain values
of x and y , J. Comput. Appl. Math. 37 (1991), 125–141.
[7] L. E ULER, Meditationes circa singulare serierum genus, Novi Commentarii academiae scientiarum
Petropolitanae 20, 1776, 140–186, E477.
[8] P. F LAJOLET AND B. S ALVY, Euler Sums and Contour Integral Representations, Experimental Math-
ematics, Vol. 7 (1998), No. 1, 15–35.
[9] C. G EORGHIOU AND A. N. P HILIPPOU, Harmonic Sums and the Zeta Function, Fibo. Quart. 21 (Feb.
1983) 29–36.
[10] P. F. J ORDAN, Infinite Sums of Psi Functions, Bulletin of the American Mathematical Society, Vol.
79, No.4, July 1973, 681–683.
[11] N. N IELSEN, Handbuch der Theorie der Gammafunktion, Leipzig, Druck und Verlag von G. B. Teub-
ner, 1906 (reprinted by Chelsea Publishing Company, Bronx, New York, 1965).
[12] S. R AMANUJAN, Manuscript Book 2 of Srinivasa Ramanujan, Tata Institute of Fundamental Research,
http://www.math.tifr.res.in/~publ/nsrBook2.pdf.
[13] G. RUTLEDGE AND R. D. D OUGLASS , Table of Definite Integrals, Amer. Math. Monthly, Vol. 45,
No. 8 (Oct., 1938), pp. 525–530.
[14] H. F. S ANDHAM, Advanced Problem 4305, Amer. Math. Monthly, Vol. 55 (7) Aug.–Sep., 1948, p.431;
solution by Martin Kneser 57(4) Apr. 1950, 267–268.
[15] R. S ITARAMACHANDRARAO, A Formula of S. Ramanujan, Journal of Number Theory 25 (1987),
1–19.
[16] A. S OFO AND D. C VIJOVI Ć, Extensions of Euler Harmonic Sums, Appl. Anal. Discrete Math. 6
(2012), 317–328, available online at http://pefmath.etf.rs.
[17] A. S OFO, General order Euler sums with rational argument, Integral Transforms Spec. Funct. (2019),
https://doi.org/10.1080/10652469.2019.1643851.
[18] A. S OFO, New classes of harmonic number identities, J. Integer Seq. 2012; 15: no. 7, Article 12.7.4,
12 pp.
[19] A. S OFO, Evaluation of integrals with hypergeometric and logarithmic functions, Open Math. 16
(2018), no. 1, 63–74.
[20] A. S OFO , H. M. S RIVASTAVA, A family of shifted harmonic sums, Ramanujan J. 37 (2015), no. 1,
89–108.
[21] G. T. W ILLIAMS , A new method of evaluating ζ (2n) , Amer. Math. Monthly 60(1) Jan. 1953, 19–25.

(Received May 25, 2019) Amrik Singh Nimbran

B3-304, Palm Grove Heights, Ardee City, Gurgaon, Haryana, India 122003
e-mail: amrikn622@gmail.com
Anthony Sofo
Victoria University
P. O. Box 14428, Melbourne City, Victoria 8001, Australia
e-mail: anthony.sofo@vu.edu.au

Journal of Classical Analysis

www.ele-math.com
jca@ele-math.com