A SERIES REPRESENTATION OF THE COTANGENT

This writeup establishes an equality of meromorphic functions,

∞  
1 X 1 1
π cot πz = + +
z n=1 z − n z + n
∞
1 X 1
= + 2z 2 − n2
.
z n=1
z

The function π cot πz (for nonintegers z ∈ C) is analytic and Z-periodic. Near

z = 0 we have
1 1
π cot πz ∼ π = ,
πz z
so that π cot πz is also meromorphic at 0, having a simple pole there with residue 1.
By Z-perodicity, the same holds at each integer n. Thus, a naı̈ve first attempt to
imitate π cot πz by a series is
X 1
.
z−n
n∈Z
However, the nth term of this series is O(1/n), so that the series is not even
summable. One can fix this problem by modifying the terms to obtain the series
 
1 X 1 1
+ + .
z z−n n
n6=0

Now the nth term is

1 1 z 1

+ = =O 2 ,
z−n n (z − n)n n
and so the new series is summable. In fact, this calculation shows that the new
series is absolutely summable, so that its terms can be rearranaged. In particular,
pairing the terms for n and −n gives
1 1 1 1 1 1
+ + − = +
z−n n z+n n z−n z+n
2z
= 2 ,
z − n2
and these are the terms of the series that we began with, in both of its forms. So
at least that series converges absolutely for any noninteger z ∈ C.
All of this said, the series that we began with (in either of its forms) is not a
Laurent series, and so part of the task here is to show that it defines a meromorphic
function at all. And even if it does, the preceding calculation has exposed a problem.
The nth term-with-correction of the series, evaluated at z + m (where m is an
integer) rather than at z, is
1 1
− .
z+m−n n
1
2 A SERIES REPRESENTATION OF THE COTANGENT

This is not any term whatsoever of the series evaluated at z. The corrections
required to make a convergent series also make a series that is not obviously Z-
periodic as a function of z, as it must be to represent the cotangent.
To show that the sum is meromorphic, recall a result from a previous writeup:
Let Ω be a region in C. Consider a sequence of differentiable functions on Ω,
{ϕ0 , ϕ1 , ϕ2 , . . . } : Ω −→ C.
Suppose that the sequence converges on Ω to a limit function
ϕ : Ω −→ C
and that the convergence is uniform on compact subsets of Ω. Then
(1) The limit function ϕ is differentiable.
(2) The sequence {ϕ0n } of derivatives converges on Ω to the derivative ϕ0 of the
limit function.
(3) This convergence is also uniform on compact subsets of Ω.

To apply the result here, let Ω = C − Z, a region in C. Define

n  
1 X 1 1
ϕn : Ω −→ C, ϕn (z) = + + , n = 0, 1, 2, . . . .
z j=1 z − j z+j

This is the sequence of partial sums of

∞  
1 X 1 1
ϕ : Ω −→ C, ϕ(z) = + + .
z j=1 z−j z+j
√
Consider any z ∈ Ω. For all j > 2|z|, the reverse triangle inequality gives
|z 2 − j 2 | ≥ j 2 − |z|2 > j 2 − j 2 /2 = j 2 /2,
and so
1 2
< 2.
z2 −j 2 j
This shows that the partial sums
n
1 X 1
ϕn (z) = + 2z
z j=1
z − j2
2

converge absolutely. Consequently, they converge to the limit function

∞
1 X 1
ϕ(z) = + 2z 2 − j2
.
z j=1
z

We need to show that the convergence is uniform on compact subsets of Ω. Let

K be such a subset, and let ε > 0 be given. There is a uniform bound b > 0 on
the absolute values |z| for all z ∈ K. Also, there a starting index n0 such that for
any n > n0 ,
∞
X 1 ε
2
< .
j=n+1
j 4b
 A SERIES REPRESENTATION OF THE COTANGENT 3

√
Consider any n such that n > n0 and also n > 2b. For such n and for all z ∈ K,
∞ ∞ ∞
X 1 X 1 X 2
|ϕ(z) − ϕn (z)| = 2z 2 − j2
≤ 2b 2 − j2
≤ 2b 2
< ε.
j=n+1
z j=n+1
z j=n+1
j

This shows that the convergence of {ϕn } to ϕ on Ω is uniform on compact subsets.

By the result, the limit function can be differentiated termwise. Now that we
no longer need the symbol n to index partial sums, we return to the more natural
notation of using it as sum-index,
∞   ∞
1 X 1 1 1 X 1
ϕ(z) = + + = + 2z 2 − n2
,
z n=1 z − n z + n z n=1
z

and
X 1
ϕ0 (z) = − .
(z − n)2
n∈Z

The second series for ϕ shows that it is odd, and the series for ϕ0 shows that it
is even. The convergence of ϕ0 is again absolute, and so ϕ0 is Z-periodic by a
calculation that rearranges terms,
X 1 X 1
ϕ0 (z + m) = − 2
=− 0 )2
where n0 = n − m.
(z + m − n) 0
(z − n
n∈Z n ∈Z

It follows that

0
ϕ(z + 1) − ϕ(z) = ϕ0 (z + 1) − ϕ0 (z) = ϕ0 (z) − ϕ0 (z) = 0,
so that
ϕ(z + 1) − ϕ(z) = c for some constant c.
To show that ϕ is Z-periodic, we need to show that c = 0. But in particular,
c = ϕ(1/2) − ϕ(−1/2) = 2ϕ(1/2) since ϕ is odd,
and so it suffices to show that ϕ(1/2) = 0. Inspect it,
∞ ∞  
X 1 X 1 1
ϕ(1/2) = 2 + 1 2
= 2 − 1 − 1 .
n=1 4
− n n=1
n − 2 n+ 2

The sum telescopes to 2, giving the desired result.

The argument so far shows that the function ϕ(z) − 1/z is also analytic at z = 0.
Therefore ϕ itself is meromorphic at 0, having a simple pole there with residue 1.
By the Z-periodicity, the same holds at each integer n. This matches the behavior
of π cot πz. Thus the difference π cot πz − ϕ(z) is entire. We want to show that it
is the zero function.
The first step is to show that the difference is bounded, making it constant by
Liouville’s theorem. Since the difference is Z-periodic in the x-direction, it suffices
to show that is bounded as |y| → ∞, and for this it suffices to show that each of
π cot πz and ϕ(z) is individually bounded as |y| → ∞. Compute first that
eπiz + e−πiz e2πiz + 1 2πi
π cot πz = πi πiz −πiz
= πi 2πiz
= πi + 2πiz .
e −e e −1 e −1
4 A SERIES REPRESENTATION OF THE COTANGENT

Also |e2πiz | = e−2πy , so limy→+∞ π cot πz = −πi and limy→−∞ π cot πz = πi. On
the other hand, suppose now that z = x + iy where 0 ≤ x < 1 and |y| > 1. Then
we have the inequalities |y| ≤ |z| ≤ |y| + 1 and
|z 2 − n2 | = |x2 − y 2 − n2 + 2ixy| ≥ y 2 + n2 − x2 ≥ y 2 + n2 − 1.
It follows that
∞
1 X 1
|ϕ(z)| ≤ + 2(|y| + 1) .
|y| n=1
y + n2 − 1
2

Let η = b|y|c. Then

∞ ∞ X η
X 1 X 1
= ,
n=1
y + n − 1 m=0 r=1 y + (mη + r)2 − 1
2 2 2

and for each m ≥ 0,

η
X 1 η 1
≤ 2 = .
r=1
y 2 + (mη + r)2 − 1 η + (mη)2 η(1 + m2 )
This shows that
∞
1 |y| + 1 X 1
|ϕ(z)| ≤ +2 ,
|y| b|y|c m=0 1 + m2
and so ϕ(z) is bounded as |y| → ∞ as well.
Thus π cot πz − ϕ(z) is constant. To see that the constant is 0, set z = 1/2.
From before, ϕ(1/2) = 0. But also π cot π/2 = 0, giving the result.
As an application, we compare the power series expansions about z = 0 of the
two now-known-to-be-equal functions
zϕ(z) and πz cot πz.
For the first expansion, compute that for |z| < 1,
∞ ∞
X 1 X 1 1
zϕ(z) = 1 + 2z 2 2 − n2
= 1 − 2z 2
·
2 1 − z 2 /n2
n=1
z n=1
n
∞ ∞ ∞ ∞
X 1 X  z 2k X X 1
= 1 − 2z 2 2
= 1 − 2 z 2k+2
2k+2
n=1
n n n=1
n
k=0 k=0
X
=1−2 ζ(k)z k .
even k≥2

That is, zϕ(z) is a generating function for the Euler–Riemann zeta function ζ(k)
at positive even values of k. On the other hand, the second expansion is essentially
a generating function for the Bernoulli numbers. Again for |z| < 1,
∞
2πiz X Bk
πz cot πz = πiz + = πiz + (2πiz)k
e2πiz − 1 k!
k=0
X (2πi)k Bk k
=1+ z .
k!
even k≥2

Comparing the two shows expansions gives Euler’s famous formula,

1 (2πi)k Bk
ζ(k) = − · for all even k ≥ 2.
2 k!
 A SERIES REPRESENTATION OF THE COTANGENT 5

In particular, this formula combines with the values B2 = 1/6, B4 = −1/30,

B6 = 1/42 to give
π2 π4 π6
ζ(2) = , ζ(4) = , ζ(6) = .
6 90 945
Euler’s formula for ζ(k) (even k ≥ 2) can also be obtained by contour integration
techniques, as in our text. The idea is that since
1 X (2πi)k Bk
π cot πz = + z k−1 ,
z k!
even k≥2

it follows that for any even k ≥ 2,

(2πi)k Bk
 
π cot πz
Resz=0 k
= .
z k!
By contour integration,
(2πi)k Bk
+ 2ζ(k) = 0,
k!
and Euler’s formula follows immediately.,
Since π cot πz is Z-periodic it also has a Fourier series expansion. This is not
the same thing as is its Laurent series expansion. Instead, if z = x + iy with y > 0
then |e2πiz | = e−2πy < 1, and so
2πi
π cot πz = πi + 2πiz
e −1
∞
X
= πi − 2πi q n , where q = e2πiz .
n=0