2 A Really Hard Infinite Series

The infinite series we will evaluate is

∞  2
X 1 1 1
S= 1 + + ··· + . (3)
n=1
n2 2 n

1 1
We adopt the shorthand notation Hn = 1 + 2 + ··· + n to denote the nth harmonic number. Recall that
Hn ∼ log n, so S clearly converges.

Step 1: Decompose S into two sums.

1
We begin by rewriting the summand. Observe that Hn+1 = Hn + n+1 . Thus,
 
1 Hn
Hn2 = Hn Hn+1 − = Hn Hn+1 −
n+1 n+1
 
1 1
= Hn Hn+1 − Hn+1 −
n+1 n+1
Hn+1 1
= Hn Hn+1 − + .
n + 1 (n + 1)2
The next step is seemingly out of nowhere. We introduce two infinite series with the eventual goal of using
summation by parts. We have
n n
! n n
!
2 Hn+1 1 X Hk X Hk X 1 X 1
Hn = Hn Hn+1 − + + − + −
n + 1 (n + 1)2 k k k2 k2
k=1 k=1 k=1 k=1
n+1
!
X Hk n+1 X 1 Xn
Hk X 1
n
= Hn Hn+1 − + + −
k k2 k k2
k=1 k=1 k=1 k=1
n+1
!
X Hk n+1 X 1 Xn
Hk X 1
n
= Hn Hn+1 − + + − .
k k2 k k2
k=2 k=2 k=1 k=1

The last equality comes from the observation that the terms corresponding to k = 1 in the first and second
summations are both 1. Continuing,
n+1 n n
!
X1 1
 X Hk X 1
2
Hn = Hn Hn+1 − Hk − + −
k k k k2
k=2 k=1 k=1
n+1 n n
!
X1 X Hk X 1
= Hn Hn+1 − · Hk−1 + −
k k k2
k=2 k=1 k=1
n+1 n n
!
X X Hk X 1
= Hn Hn+1 − (Hk − Hk−1 ) Hk−1 + −
k k2
k=2 k=1 k=1
n+1 n+1 n n
!
X X
2
X Hk X 1
= Hn Hn+1 − Hk Hk−1 + Hk−1 + −
k k2
k=2 k=2 k=1 k=1
n+2 n+1 n n
!
X X
2
X Hk X 1
= Hn Hn+1 − Hk−1 Hk−2 + Hk−1 + −
k k2
k=3 k=2 k=1 k=1
n+2 n+1 n n
!
X X
2
X Hk X 1
= Hn Hn+1 − Hk−1 Hk−2 + Hk−1 + −
k k2
k=3 k=2 k=1 k=1
n+1 n n
!
X
2

X Hk X 1
=1+ Hk−1 − Hk−1 Hk−2 + − .
k k2
k=3 k=1 k=1

6
The last equality comes from picking off the last term of the first sum and the first term of the second sum.
Simplifying,
n+1 n n
!
2
X X Hk X 1
Hn = 1 + Hk−1 (Hk−1 − Hk−2 ) + −
k k2
k=3 k=1 k=1
n+1 n n
!
X Hk−1 X Hk X 1
=1+ + −
k−1 k k2
k=3 k=1 k=1
n+1 n n
!
X Hk−1 X Hk X 1
= + −
k−1 k k2
k=2 k=1 k=1
n n
X Hk X 1
=2 − .
k k2
k=1 k=1

All of this shows that

∞ ∞ n ∞ n
X Hn2 X 1 X Hk X 1 X 1
S= =2 − . (4)
n=1
n2 n=1
n2 k n=1
n2 k2
k=1 k=1
| {z } | {z }
S1 S2

Step 2: Simplify S1 .
Next we consider S1 . Note that
k k Z
Hk 1X1 1 X 1 j−1
= = x dx
k k j=1 j k j=1 0
Z 1 k
1 X
= xj−1 dx
k 0 j=1
1
1 − xk
Z
1
= dx
k 0 1−x

where we have used the partial sum formula for a geometric series in the last equality. Continuing,
1 1 1
1 − xk
Z Z Z
Hk 1 1
= dx = tk−1 dt dx.
k k 0 1−x 0 1−x x

Next, we change the order of integration, taking care to change the bounds appropriately.
Z 1 Z t Z 1
Hk 1
= tk−1 dx dt = − tk−1 log(1 − t) dt.
k 0 0 1−x 0

We continue simplifying S1 in the same manner. In particular,

n n Z 1
X Hk X
=− tk−1 log(1 − t) dt
k 0
k=1 k=1
Z 1
1 − tn
=− log(1 − t) dt
0 1−t
Z 1
log(1 − t) 1 n−1
Z
= −n y dy dt
0 1−t t
Z 1 Z y
log(1 − t)
= −n y n−1 dt dy.
0 0 1−t

7
The inner integral can be evaluated by a simple substitution to yield
n Z 1
X Hk n
= y n−1 log2 (1 − y) dy.
k 2 0
k=1

So
∞ n ∞
1 X 1 1 n−1
Z
X 1 X Hk
2
= y log2 (1 − y) dy
n=1
n k 2 n=1
n 0
k=1
∞
Z 1 !
1 1 X yn
= log2 (1 − y) dy.
2 0 y n=1 n
P∞ yn
Recall that log(1 − y) = − n=1 n for −1 ≤ y < 1. Thus,
∞ n
1 1 log3 (1 − y)
Z
X 1 X Hk
= − dy.
n=1
n2 k 2 0 y
k=1

Now we make the change of variables z = 1 − y. This yields

∞ ∞ Z
!
1
log3 (1 − y) 1
log3 z 1
1X 1 n
Z Z Z
1 1 1 3
X
− dy = − dz = − log z z n
dz = − z log3 z dz.
2 0 y 2 0 1−z 2 0 n=0
2 n=0 0

Next, we integrate by parts three times. First, an observation. By L’Hopital’s rule,

logm z
lim 1 =0
z→0
zj

for any m, j > 0. Thus, if we let u = log3 z and dv = z n dz and integrate by parts, the boundary terms
disappear and we have
Z 1 Z 1
3
z n log3 z dz = − z n log2 z dz.
0 n+1 0
Integrating by parts two more times yields
Z 1 Z 1 Z 1
6 6 6
z n log3 z dz = z n log z dz = − z n dz = − .
0 (n + 1)2 0 (n + 1)3 0 (n + 1)4

Putting all of this together yields

∞ n ∞ Z ∞ ∞
X 1 X Hk 1X 1 n 3
X 1 X 1
S1 = 2
= − z log z dz = 3 4
= 3 .
n=1
n k 2 n=0 0 n=0
(n + 1) n=1
n4
k=1

Step 3: Simplify S2 .
(2) 1 1
Next, we consider S2 . We adopt the shorthand notation Hn = 1+ 22 + ··· + n2 to denote the nth
harmonic number of order 2. Then
∞ (2)
X Hn
S2 = .
n=1
n2
We proceed by summation by parts. We have
N N  N N
X 1 (2) X
(2) (2)

(2)
X
(2) (2)
X (2)
H
2 n
= 1 + Hn − Hn−1 Hn = 1 + H n H n − Hn−1 Hn(2) .
n=1
n n=2 n=2 n=2

8
Next, we reindex the first sum and pull off terms from both in order to simplify.
N N   N N
X X (2) (2) (2) 1 X (2) (2)
X (2)
1+ Hn(2) Hn(2) − Hn−1 Hn(2) =1+ HN HN − 1+ + Hn−1 Hn−1 − Hn−1 Hn(2)
n=2 n=2
4 n=3 n=3
2 N
1 (2)
 X (2)

(2)

= − + HN + Hn−1 Hn−1 − Hn(2)
4 n=3
2 N
1 (2)
 X (2) 1
= − + HN − Hn−1 ·
4 n=3
n2
2 N

(2)
X (2) 1
= HN − Hn−1 · .
n=2
n2

(2)
H1
The last inequality comes from the observation that 22 = 14 . Continuing,

N (2) N (2) 1 N (2) N


(2)
2 X Hn−1 
(2)
2 X Hn − n2

(2)
2 X Hn X 1
HN − = H N − = HN − + .
n=2
n2 n=2
n2 n=1
n 2
n=1
n 4

Putting all of this together and sending N → ∞ yields

∞
!2 ∞
X 1 X 1
S2 = 2
− S2 + 4
n=1
n n=1
n

and so !2
∞ ∞
1 X 1 1X 1
S2 = + .
2 n=1
n2 2 n=1 n4
P∞ 1
P∞ 1
Step 4: Algebraically relate n=1 n4 to n=1 n2 .
P∞ 1
There are many ways to evaluate n=1 n4 .
Typical arguments use Fourier series, complex analy-
sis, or infinite product expansions. In the spirit of presenting a complete solution which only uses inte-
P∞
grals, infinite sums, and ordinary power series, we present a pure algebraic manipulation of n=1 n14 to
P∞ 1
n=1 n2 . This technique comes from https://math.stackexchange.com/q/1006510 and the refer-
ence therein. Note that
∞ ∞  
X 1 1X 2 1 2
= + 2 2+ 3 .
n=1
n4 5 n=1 n · n3 n ·n n ·n

Let
2 1 2
am,n = 3
+ 2 2+ 3 .
mn m n m n
Observe1 that
∞
X X X X
an,n = am,n − am,n − am,n
n=1 m,n≥1 n>m≥1 m>n≥1
X X X
= am,n − am,m+n − am+n,n
m,n≥1 m,n≥1 m,n≥1
X
= (am,n − am,m+n − am+n,n ) .
m,n≥1

1 Note that there are some divergent double sums here. It’s probably okay though!

9
Now we simplify the above double sequence.
 
2 1 2
am,n − am,m+n − am+n,n = + 2 2+ 3
mn3 m n m n
   
2 1 2 2 1 2
− + 2 + 3 − + + .
m(m + n)3 m (m + n)2 m (m + n) (m + n)n3 (m + n)2 n2 (m + n)3 n
| {z }
A

Isolating and simplifying A,

 2 3
2m n + mn3 (m + n) + 2n3 (m + n)2 + 2m3 n2 + m3 n(m + n) + 2m3 (m + n)2

A=−
m3 n3 (m + n)3
 2 2
2m n (m + n) + mn3 (m + n) + 2n3 (m + n)2 + m3 n(m + n) + 2m3 (m + n)2

=−
m3 n3 (m + n)3
 2 2
2m n + mn + 2n (m + n) + m3 n + 2m3 (m + n)
3 3

=−
m3 n3 (m + n)2
mn(2mn + n + m2 ) + 2n3 (m + n) + 2m3 (m + n)
2
 
=−
m3 n3 (m + n)2
mn(n + m) + 2n (m + n) + 2m3 (m + n)
2 3
 
=−
m3 n3 (m + n)2
mn(n + m) + 2n3 + 2m3
 
=−
m3 n3 (m + n)

So
2m2 + mn + 2n2 mn(n + m) + 2n3 + 2m3
   
am,n − am,m+n − am+n,n = −
m3 n3 m3 n3 (m + n)
2m2 (n + m) + 2n2 (n + m) − 2n3 − 2m3
=
m3 n3 (m + n)
2m2 n + 2n2 m
=
m3 n3 (m + n)
2m + 2n
=
m n2 (m + n)
2

2
= .
m2 n2
Putting everything together, we have
∞ ∞
X 1 1X 1 X
4
= an,n = (am,n − am,m+n − am+n,n )
n=1
n 5 n=1 5
m,n≥1
1 X 2
=
5 m n2
2
m,n≥1
∞ X∞
2 X 1 1
= ·
5 m=1 n=1 m2 n2
∞
!2
2 X 1
= .
5 n=1 n2

P∞ 1
Step 5: Evaluate n=1 n2 .

10
 P∞
Again, there are numerous ways to evaluate n=1 n12 . We present a solution due to [1] which is themat-
ically consistent with everything else done thus far. Note that
∞ ∞ ∞ ∞ ∞
X 1 X 1 X 1 1X 1 X 1
2
= 2
+ 2
= 2
+ .
n=1
n n=1
(2n) n=0
(2n + 1) 4 n=1 n n=0
(2n + 1)2
Solving for the desired sum in the above equation gives
∞ ∞
X 1 4X 1
2
= .
n=1
n 3 n=0
(2n + 1)2
We evaluate the latter sum by converting it to a double integral. Note that
∞ ∞ Z 1  Z 1 
X 1 X
2n 2n
= x dx y dy
n=0
(2n + 1)2 n=0 0 0
Z 1Z 1X ∞
= (x2 y 2 )n dx dy
0 0 n=0
Z 1 Z1
1
= dx dy.
0 0 1 − x2 y 2
Next, we make the change of variables
sin u sin v
x= and y= .
cos v cos u
Note that " # " #
 ∂ x cos u sin u sin v 
 u ∂v x  
 = 1 − tan2 u tan2 v.
2
 =  sincos v cos v
  

 ∂u y u sin v cos v
∂v y   cos 2u cos u

It follows that dx dy = (1 − tan2 u tan2 v) du dv. Let E be the region in the the square 0 ≤ u, v ≤ π2 which
is the image of 0 ≤ x, y ≤ 1 under this transformation. Note that this region is defined by sin u ≤ cos v
and sin v ≤ cos u. Note that equality occurs in both inequalities if v = π2 − u. This line divides the square
0 ≤ u, v ≤ π2 into two triangles. The inequalities dictate that E is the triangle with vertices (0, 0), (π/2, 0),
and (0, π/2). Therefore,
Z 1Z 1
1 − tan2 u tan2 v π2
ZZ ZZ
1 1 π π
2 2
dx dy = 2 2 du dv = du dv = · · = .
0 0 1−x y E 1 − tan u tan v E 2 2 2 8
Thus,
∞ ∞
X 1 4X 1 4 π2 π2
2
= 2
= · = .
n=1
n 3 n=0 (2n + 1) 3 8 6

Step 6: Put it all together.

Summarizing steps 1 through 5 gives the final answer:
∞ ∞
!2 ∞ ∞ ∞
!2
X 1 1 X 1 1X 1 11 X 1 1 X 1
S = 2S1 − S2 = 6 4
− 2
− 4
= 4
−
n=1
n 2 n=1 n 2 n=1 n 2 n=1 n 2 n=1
n2
∞
!2 ∞
!2
11 2 X 1 1 X 1
= · −
2 5 n=1 n2 2 n=1 n2
 2
17 π 2
= .
10 6
Thus,
∞ 2
17π 4

X 1 1 1
1 + + · · · + = .
n=1
n2 2 n 360

11