You are on page 1of 14

# SUMS OF ARCTANGENTS AND SOME FORMULAE OF

RAMANUJAN
GEORGE BOROS AND VICTOR H. MOLL

## Abstract. We present diverse methods to evaluate arctangent and related

sums.

1. Introduction
The evaluation of arctangent sums of the form
(1.1)

tan1 h(k)

k=1

for a rational function h reappear in the literature from time to time. For instance
the evaluation of

X
2
3
(1.2)
tan1 2 =
k
4
k=1

## is completely elementary and was proposed by J. Anglesio [1] in 1993. This is a

classical problem that appears in [7, 9, 13], among other places. The evaluation of

X
1 1
1 tan(/ 2) tanh(/ 2)

tan
(1.3)
= tan
k2
tan(/ 2) + tanh(/ 2)
k=1
is somewhat less elementary. R.J. Chapman [6] proposed in 1990 to evaluate this
sum in closed form (in terms of a finite number of trigonometric and hyperbolic
functions). This was solved by A. Sarkar [15] using the techniques described in
Section 3.
The goal of this paper is to discuss the evaluation of these sums and the related
rational sums
n
X
(1.4)
S(n) =
R(k)
k=1

## We make use of the addition formula for tan1 x:

(
x+y
tan1 1xy
1
1
(1.5)
tan x + tan y =
x+y
tan1 1xy
+ sign x
Date: March 23, 2001.
1991 Mathematics Subject Classification. Primary 33.
Key words and phrases. Arctangent, Series.
1

if xy < 1
if xy > 1

and
tan1 x + tan1

(1.6)

1
x

sign x.
2

## 2. The method of telescoping

The closed-form evaluation of a finite sum
n
X
S(n) :=
ak
k=1

ak

= bk bk1 .

S(n) :=

n
X

ak =

k=1

n
X

bk bk1 = bn b0 .

k=1

## This method can be extended to situations in which the telescoping nature of ak is

hidden by a function.
Theorem 2.1. Let f (x) be of fixed sign and define h by
h(x)

(2.1)

f (x + 1) f (x)
.
1 + f (x + 1)f (x)

Then
n
X

(2.2)

tan1 h(k) =

k=1

(2.3)

tan1 h(k) =

k=1

Proof. Since
tan1 h(k)

## (2.2) follows by telescoping.

Note. The hypothesis on the sign of f is included in order to avoid the case xy > 1
in (1.5). In general, (2.2) has to be replaced by
(2.4)

n
X

k=1

tan1 h(k)

## tan1 f (n) tan1 f (1) +

sign f (k),

where the sum is taken over all k between 1 and n for which f (k)f (k + 1) < 1.
Thus (2.2) is always correct up to an integral multiple of . The restrictions on
the parameters in the examples described in this paper have the intent of keeping
f (k), k N of fixed sign.
Example 2.1. Let f (x) = ax + b, where a, b are such that f (x) 0 for x 1.
Then
a
h(x) =
(2.5)
a2 x2 + a(a + 2b)x + (1 + ab + b2 )

## and (2.3) yields

(2.6)

tan1

k=1

a
a2 k 2 + a(a + 2b)k + (1 + ab + b2 )

tan1 (a + b).
2

Special cases: a = 1 and b = 0 give f (x) = x and h(x) = 1/(x2 + x + 1), and the
sum is

1
(2.7)
=
.
tan1 2
k +k+1
4
k=1

(2.8)

tan1

k=0

2
(2k + 1)2

.
2

## Differentiating (2.6) with respect to a yields

X
pa,b (k)

(2.9)

k=1

qa,b (k)

1
,
1 + (a + b)2

where
= a2 k 2 + a2 k (1 + b2 )

pa,b (k)
and
qa,b (k)

## = a4 k 4 + 2a3 (a + 2b)k 3 + a2 (2 + a2 + 6ab + 6b2 )k 2 +

2a(a + 2b)(1 + ab + b2 )k + (1 + b2 )(1 + a2 + 2ab + b2 ).

## Special cases: a = 1 and b = 0 give

X
k2 + k 1
4
k + 2k 3 + 3k 2 + 2k + 2

k=1

1
.
2

## a = 1/2 and b = 1/3 give

k=1

81k 4

9k 2 + 9k 40
+ 1269k 2 + 1932k + 2440

378k 3

1
.
61

## Let a = 2b and differentiate with respect to b to produce

X
4b2 k 2 (1 + b2 )
1
=
.
16b4 k 4 + 8b2 (1 b2 )k 2 + (1 + b2 )2
2(1 + b2 )
k=1

## Mathematica evaluates this last sum.

Example 2.2. This example considers the quadratic function f (x) = ax2 + bx + c
under the assumption that f (k), k N has fixed sign. For instance this happens if
b2 4ac 0. Define
1 + ac + bc + c2
ab + b2 + 2ac + 2bc
a2 + 3ab + b2 + 2ac
2a(a + b)

a0
a1
a2
a3

:=
:=
:=
:=

a4

:= a2

## GEORGE BOROS AND VICTOR H. MOLL

Then
(2.10)

h(x)

and thus

tan1

k=1

2ax + a + b
a4 x4 + a3 x3 + a2 x2 + a1 x + a0

2ak + a + b
a4 k 4 + a3 k 3 + a2 k 2 + a1 k + a0

tan1 (a + b + c).
2

## Special cases: b = a, c = a/2 yields

tan1

k=1

8ak
4a2 k 4 + (a2 + 4)

a
tan1 .
2
2

(2.11)

tan1

k=1

8k
4k 4 + 5

1
tan1 .
2
2

## b = a, c = 0 yields the sum

(2.12)

tan1

k=1

a2 k 4

2ak
a2 k 2 + 1

,
2

which is independent of a.
Additional examples can be given by telescoping twice (or even more).
Corollary 2.2. Let f and h be related by
h(x)

(2.13)
Then
n
X

tan1 h(k)

f (x + 1) f (x 1)
.
1 + f (x + 1)f (x 1)

k=1

In particular,

(2.14)

tan1 h(k)

k=1

tan1 h(k)

tan1 h(k)

## tan1 f (k + 1) tan1 f (k 1).

Thus
n
X

k=1

=
=

n
X
 1

tan f (k + 1) tan1 f (k 1)

k=1

n
n
X
 1
 X
 1

tan f (k + 1) tan1 f (k) +
tan f (k) tan1 f (k 1)

k=1
1

tan

k=1

X
2
(2.15)
tan1 2
k

3
4

k=1

## corresponds to f (k) = k so that h(k) = 2/k 2 . This is the problem proposed by

Anglesio [1].
Example 2.4. Take f (k) = 2/k 2 so that h(k) = 8k/(k 4 2k 2 + 5). It follows
that

X
1
8k
= tan1 .
tan1 4
k 2k 2 + 5
2
k=1

## This sum is part b) of the Problem proposed in [1].

Example 2.5. Take f (k) = a/(k 2 + 1). Then h(k) = 4ak/(k 4 + a2 + 4), so that

X
4ak
a
(2.16)
tan1 4
= tan1 + tan1 a.
k + a2 + 4
2
k=1

X
(2.17)
tan1
k=1

a=

4k
+5

k4

2 gives

(2.18)

tan

k=1

1
+ tan1 .
4
2

4 2k
k4 + 6

.
2

## Differentiating (2.16) with respect to a gives

X
4k(k 4 + 4 a2 )
8
2
k + 2(a + 4)k 4 + 16a2 k 2 + (a4 + 8a2 + 16)

k=1

(a2

3(a2 + 2)
.
+ 1)(a2 + 4)

## The special case a = 0 yields

(2.19)

k=1

k
k4 + 4

3
8

and a = 2 gives
(2.20)

k=1

k5
k 8 + 16k 4 + 64k 2 + 64

9
.
80

## Mathematica evaluates the first sum but not the second.

The examples described above are rather artficial. The interesting question is
to find f (x) given the function h. In general it is not possible to find f in closed
form. In the case of Example 2.3, we need to solve the functional equation
(2.21)

## 2 [1 + f (x 1)f (x + 1)] = x2 [f (x + 1) f (x 1)] .

A polynomial solution of (2.21) must have degree at most 2 and trying f (x) =
ax2 + bx + c yields f (x) = x.

## 3. The method of zeros

A different technique for the evaluation of arctangent sums is based on the factorization of the product
(3.1)

pn

n
Y

:=

(ak + ibk )

k=1

Arg(pn )

n
X

tan1

k=1

bk
.
ak

(3.2)

pn (z)

n
Y

(z zk )

k=1

(3.3)

Arg(pn (z))

n
X

tan1

k=1

x xk
.
y yk

obtain
n
X
x cos(2k/n)
Arg(z n 1) =
tan1
(3.4)
.
y sin(2k/n)
k=1

(3.5)

sin z

= z


Y

k=1

z2
k2

(3.6)

tan1

k=1

2xy
k 2 x2 + y 2

= tan1

y
tanh y
tan1
.
x
tan x

## Special cases: x = y yields

(3.7)

tan1

k=1

2x2
k2

x = y = 1/ 2 gives
(3.8)

tan1

k=1

1
k2

tanh x
tan1
.
4
tan x

tanh (/ 2)
,
tan1
4
tan(/ 2)

## which corresponds to (1.3).

x = y = 1/2 yields
(3.9)

k=1

tan1

1
2k 2

.
4

## Differentiating (3.7) gives

X
k2
(3.10)
4
k + 4x4

sin 2x sinh 2x
.
4x cos 2x cosh 2x

k=1

In particular, x = 1 yields

(3.11)

k=1

k2
+4

coth .
4

k4

## The identity (3.10) is comparable to Ramanujans evaluation

X
k2

sinh x 3 3 sin x

(3.12)
=
k 4 + x2 k 2 + x4
2x 3 cosh x 3 cos x
k=1

## discussed in [3], Entry 4 of Chapter 14.

Glasser and Klamkin [10] present other examples of this technique.
4. A functional equation
The table of sums and integrals [11] contains a small number of examples of
finite sums that involve trigonometric functions of multiple angles. In Section 1.36
we find
n
X
x
x
(4.1)
22k sin4 k = 22n sin2 n sin2 x
2
2
k=1
n
X

(4.2)

k=1

x
1
sec2 k
2k
2
2

cosec2 x

1
x
cosec2 n
2n
2
2

## and Section 1.37 consists entirely of the two sums

n
X
1
x
1
x
(4.3)
tan k =
cot n 2 cot 2x
2k
2
2n
2
k=0

(4.4)

n
X
1
x
tan2 k
22k
2

k=0

22n+2 1
1
x
+ 4 cot2 2x 2n cot2 n .
3 22n1
2
2

Theorem 4.1. Let
(4.5)

F (x) =

f (x, k)

and

G(x) =

k=1

k=1

(4.6)

F (x)

(2) F (

(n)

(x))

n1
X

j=0

F (x) + G(x) = 2

f (x, 2k) = 2

k=1

## Repeat this argument to obtain the result.

k=1

f ((x), k) = 2F ((x)).

## GEORGE BOROS AND VICTOR H. MOLL

Example 4.1. Let f (x, k) = 1/(x2 + k 2 ), so that = 1/4 and (x) = x/2. Since
F (x) =

k=1

x2

1
x coth x 1
=
2
+k
2x2

and

X
(1)k
x csch x 1
G(x) =
=
,
x2 + k 2
2x2
k=1

(4.7)

j=0

x
2j
sinh 2j x

x
.
tanh x

X
2j coth 2j
2j sinh 2j
j=0

(4.8)

1 + 4e2 e4
.
1 2e2 + e4

## Now go back to (4.7), replace x by ln t, differentiate with respect to t, set t = ae,

differentiate with respect to a, and set a = e to produce
(4.9)

X
2 22j + csch2 2j sech2 2j
22j sinh21j
j=0

.
e12 3e8 + 3e4 1

(4.10)

F (x) =

x
X
f
k

and

G(x) =

k=1

(1)k f

k=1

x
k

(4.11)

F (x)

= 2n F (2n x) +

n
X

2k G(2k x).

k=1

## In particular, if F is bounded, then

(4.12)

F (x)

2k G(2k x).

k=1

Proof. The function f (x/k) satisfies the conditions of Theorem 4.1 with = 1 and
(x) = x/2. Thus
F (x) = 2n F (x/2n )

n1
X

2j G(x/2j ).

j=1

## Now replace x by x/2 to obtain (4.11). Finally let n to obtain (4.12).

The key to the proof of Theorem 4.1 is the identity F (x) + G(x) = 2F ((x)).
We now present an extension of this result.

## Theorem 4.3. Let F, G be functions that satisfy

(4.13)

F (x)

= r1 F (m1 x) + r2 G(m2 x)

## for some parameters r1 , r2 , m1 , m2 . Then

(4.14)

r2

n
X

r1k1 G(mk1
m2 x) = F (x) r1n F (mn1 x).
1

k=1

F (m1 x)

F (x)

## Formula (4.14) follows by induction.

We now present two examples that illustrate Theorem 4.3. These sums appear
as entries in Ramanujans Notebooks.
Example 4.2. The identity
(4.15)

cot x =

1
x 1
x
cot tan
2
2 2
2

shows that F (x) = cot x, G(x) = tan x satisfy (4.13) with r1 = 1/2, r2 = 1/2 and
m1 = m2 = 1/2. We conclude that
(4.16)

n
X

2k tan

k=1

x
2k

1
x
cot n cot x.
2n
2

This is (4.3). It also appears as Entry 24, p. 364, of Ramanujans Third Notebook
as described in Berndt [4], p. 396. Similarly, the identity
(4.17)

sin2 (2x) =

4 sin2 x 4 sin4 x

yields (4.1). The reader is invited to produce proofs of (4.2) and (4.4) in the style
presented here.

cot x =

(4.18)

cot

x
csc x
2

## satisfies (4.13) with F (x) = cot x, G(x) = csc x and parameters r1 = 1, r2 =

1, m1 = 1/2, m2 = 1. We obtain
(4.19)

n
X

k=1

csc

x
2k1

cot

x
cot x.
2n

[4], p. 398.

10

## GEORGE BOROS AND VICTOR H. MOLL

Example 4.4. The application of Theorem 4.1, or Corollary 4.2, requires an analytic expression for F and G. A source of such expressions is Jolley [12]. Indeed,
entries 578 and 579 are

tan1

2x2
k2

(1)k1 tan1

2x2
k2

(4.20)

k=1

(4.21)

k=1

tanh x
tan1
4
tan x

sinh x
+ tan1
.
4
sin x

These results also appear in [5], page 314. Applying one step of Proposition 4.2 we
conclude that
tanh x
tanh 2x
sinh 2x
(4.22)
2 tan1
= tan1
+ tan1
.
tan x
tan 2x
sin 2x
We also obtain
n
n
X
sinh 2k x
1 tanh x
n
1 tanh 2 x
(4.23)
2k tan1
=
tan
,

2
tan
sin 2k x
tan x
tan 2n x
k=1

X
sinh 2k x
tanh x
(4.24)
= tan1
.
2k tan1
k
sin 2 x
tan x
k=1

## Differentiating (4.23) gives

2

n
X
cos 2k x sinh 2k x cosh 2k x sin 2k x

k=1

## cos 2k+1 x cosh 2k+1 x

sin 2x sinh 2x
sin 2n+1 x sinh 2n+1 x
+
.
cos 2x cosh 2x cos 2n+1 x cosh 2n+1 x
Letting n and using the identity
=

(4.25)


cos 2k+1 x cosh 2k+1 x = 2 sin2 2k x + sinh2 2k x

yields

X
cosh 2k x sin 2k x sinh 2k x cos 2k x
k=1

sin2 2k x + sinh2 2k x

+ sign x.
tan2 x + tanh2 x

## For example, x = gives

(4.26)

csch 2k

coth 1.

k=1

5. Reduction to telescoping
The sum
(5.1)

S(f )

n
X

f (k, t),

k=1

## depending on the parameter t, is said to telescope if the summand can be written

in the form
(5.2)

f (k, t) = f1 (k, t) f1 (k + m, t)

11

## for a function f1 and m Z fixed. In this section we discuss a method to determine

if the sum S telescopes.
We first consider the case of a function f (k, t) that is rational in k. Then the
question of telescoping is decided by examining the partial fraction decomposition
of f . For simplicity, we assume that the poles of f are simple and we omit the
parameter t.
Proposition 5.1. Let
f (k)

(5.3)

r
X

j=1

aj
k kj

## be the partial fraction decomposition of f . Then S(f ) telescopes if and only if f

can be written as

s 
X
bj
bj
(5.4)
f (k) =

k kj
k kj mj
j=1
where mj Z.
We now observe that if (5.1) telescopes, then so do the sums
n
X

f (k, t)
k

(5.5)

and

k=1

n
X

f (k, t).
t

k=1

n
X
(5.6)
tan1 R(k)
S(n) =
k=1

n
X
1
R
(5.7)
S1 (n) =
,
1 + R2 (k) k
k=1

## by differentiating with respect to k. For instance, the sum in

rational function with poles at
p
p
p
1
1
1
(1 1 4i/a), (1 + 1 4i/a), (1 1 + 4i/a),
2
2
2
and these can be paired as required in Proposition 5.1.

(2.12) produces a
p
1
(1 + 1 + 4i/a),
2

## Example 5.1. Consider the sum



X
2k 1
2k 2 2k + 1
2
1
S =
k tan1 4

(k
+
1)tan
.
k 2k 3 + k 2 1
k 4 2k 3 + k 2 + 1
k=1

To evaluate S we introduce
2k 2 2k + 1
2k 1
A(k) = k tan1 4
(k 2 + 1)tan1 4
3
2
k 2k + k 1
k 2k 3 + k 2 + 1
and observe that
d3
(5.8)
A(k) = B(k) B(k 1),
dk 3
where
6
8(3k 3 5k 8) 64(k 3 k 1)
(5.9)
B(k) = 4

+
.
k +1
(k 4 + 1)2
(k 4 + 1)3

12

## Let C(k) be the function obtained by integrating B(k) three times:

(5.10)

C(k)

= (k 2 + k + 1)tan1 k 2 + P2 (k)

## where P2 (k) is a polynomial of degree 2.

Integrating backwards we obtain

N 
X
2k 2 2k + 1
2k 1
1
2
1
k tan
(k + 1) tan
k 4 2k 3 + k 2 1
k 4 2k 3 + k 2 + 1
k=2

= C(N ) C(1)
= (N 2 + N + 1)tan1 N 2 + Q2 (N ),
where Q2 (N ) = P2 (N ) C(1) is another polynomial of degree 2.We have started
the sum at k = 2 because k 4 2k 3 + k 2 1 has a zero at k = 12 ( 5 + 1) 1.618.
The polynomial Q2 (N ) = aN 2 + bN + c can be determined from
aN 2 + bN + c =

(5.11)

N
X

A(k) + (N 2 + N + 1)tan1 N 2

k=2

## by evaluating (5.11) at three distinct values of N . The result is



(5.12)
Q2 (N ) =
2N 2 + 2N 1 .
4
Thus
N 
X
k tan1

k=2

2k 2 2k + 1
2k 1
(k 2 + 1) tan1 4
4
k 2k 3 + k 2 1
k 2k 3 + k 2 + 1

= (N 2 + N + 1)tan1 N 2 +
so that

X
k tan1
k=1

(2N 2 + 2N 1),
4


2k 2 2k + 1
2k 1
3
2
1
(k + 1) tan
=1
.
4
3
2
4
3
2
k 2k + k 1
k 2k + k + 1
2

Note. Naturally this idea applies to a more general class of sums. For instance,
sums of the form
(5.13)

L(n)

n
X

p(k) ln R(k)

k=1

and
(5.14)

A(n) =

n
X

k=1

## for a polynomial p, can be reduced to a sum with rational summand by succesive

differentiation.

13

6. A dynamical system
In this section we describe a dynamical system that appears in the evaluation of
arctangent sums. Define
n
n
X
X
1
tan1 .
tan1 k and yn = tan
xn = tan
k
k=1

k=1

Then x1 = y1 = 1 and
xn =

nyn1 + 1
xn1 + n
and yn =
.
1 nxn1
n yn1

## Proposition 6.1. Let n N. Then

(
yn
(6.1)
xn =
1/yn

if n is even
if n is odd

that is
(6.2)

tan

n
X

tan1 k

= tan

k=1

n
X

tan1

k=1

1
k

if n is even and
(6.3)

tan

n
X

tan1 k

cotg

k=1

n
X

tan1

k=1

1
k

if n is odd.
Proof. The recurrence formulas for xn and yn can be used to prove the result
directly. A pure trigonometric proof is presented next. If n is even then
tan

2m
X

k=1

tan1 k + tan

2m
X

k=1

tan1

1
k

tan

2m
X

tan1 k + tan

k=1

tan

2m
X

/2 tan1 k

k=1
1

tan

k=1

2m
X

k + tan m

2m
X

tan

k=1

0.

## A similar argument holds for n odd.

This dynamical system suggests many interesting questions. We conclude by
proposing one of them: Observe that x3 = 0. Does this ever happen again?
References
[1] ANGLESIO, J.: Elementary problem 10292. Amer. Math. Monthly 100, 1993, 291. Solution
by several authors Amer. Math. Monthly, 103, 270-272.
[2] BERNDT, B.C.: Ramanujans Notebooks, Part I. Springer-Verlag, New York, 1985.
[3] BERNDT, B.C.: Ramanujans Notebooks, Part II. Springer-Verlag, New York, 1989.
[4] BERNDT, B.C.: Ramanujans Notebooks, Part IV. Springer-Verlag, New York, 1994.
[5] BROMWICH, T.J.: An introduction to the theory of infinite series. 1926.
[6] CHAPMAN, R.J.: Elementary problem 3375. Amer. Math. Monthly 97, 1990, 239-240.
[7] CHRYSTAL, G.: Algebra, Part II, 2nd. ed., A. and C. Black, London, 1922, p. 357.
[8] DARLING, D.: Telescoping series of arctangents. Amer. Math. Monthly 103, 1996, 270-272.
[9] GLAISHER, J.W.L.: A theorem in trigonometry, Quart. J. Math. 15, 1878, 151-157.
[10] GLASSER, M.L. - KLAMKIN, M.S.: On some inverse tangent summations. Fibonacci Quarterly, 14, 1976, 385-388.

14

## GEORGE BOROS AND VICTOR H. MOLL

[11] GRADSHTEYN, I.S. - RYZHIK, I.M.: Table of Integrals, Series and Products. Sixth Edition,
ed. Alan Jeffrey, assoc. ed. D. Zwillinger. Academic Press, 2000.
[12] JOLLEY, L.B.W.: Summation of series, 1961
[13] LONEY, S.L.: Plane Trigonometry, Part II, Cambridge University Press, Cambridge, 1893,
p. 206.
[14] MILLER, A. - SRIVASTAVA, H.M.: On Glaishers infinite sums involving the inverse tangent function. Fibonacci Quart., 1992, 290-294.
[15] SARKAR, A.: The sum of arctangents of reciprocal squares. Amer. Math. Monthly, 98, 1990,
652-653.
[16] SCHAUMBERGER, N.: An old arctangent series reappears. The College Math. J. Problem
399. May 1990, vol. 21, 253-254.
Department of Mathematics, Xavier University, New Orleans, LA 70125