Writing π as sum of arcotangents with linear

recurrent sequences, Golden mean and Lucas

numbers
arXiv:1409.6455v1 [math.NT] 23 Sep 2014

Marco Abrate, Stefano Barbero, Umberto Cerruti, Nadir Murru

Abstract
In this paper, we study the representation of π as sum of arcotan-
gents. In particular, we obtain new identities by using linear recurrent
sequences. Moreover, we provide a method in order to express π as
sum of arcotangents involving the Golden mean, the Lucas numbers,
and more in general any quadratic irrationality.

1 Expressions of π via arctangent function with

linear recurrent sequences
The problem of expressing π as the sum of arctangents has been deeply
studied during the years. The first expressions are due to Newton (1676),
Machin (1706), Euler (1755), who expressed π using the following identities
! ! !
π 1 4 1
= 2 arctan + arctan + arctan
2 2 7 8
! !
π 1 1
= arctan + arctan
4 2 3
! !
π 1 3
= 5 arctan + 2 arctan ,
4 7 79
respectively (see, e.g., [12] and [13]). Many other identities and methods to
express and calculate π involving the arctangent function have been devel-
oped. Some recent results are obtained in [6] and [2].
In this section, we find a method to generate new expressions of π in
terms of sum of arctangents, mainly using the properties of linear recurrent
sequences. For the sake of simplicity, we will use the following notation:

A(x) = arctan(x).

1
 1
It is well–known that for x, y ≥ 0, if y 6=
x
(
A(x ⊙ y) if xy < 1,
A(x) + A(y) =
A(x ⊙ y) + sign(x)π if xy > 1,

where
x+y
x⊙y = .
1 − xy
Let us denote by x⊙n the n–th power of x with respect to the product ⊙.

Remark 1. The product ⊙ is associative, commutative and 0 is the identity.

Definition 1. We denote by a = (an )+∞ n=0 = W(α, β, p, q) the linear recur-

rent sequence of order 2 with characteristic polynomial t2 − pt + q and initial
conditions α and β, i.e.,

a0 = α

a1 = β

an = pan−1 − qan−2 ∀n ≥ 2 .


Theorem 1. Given n ∈ N and x ∈ R, with x 6= ±1, we have

!⊙n
1 vn (x)
= , ∀n ≥ 1
x un (x)

where
2 2
(un (x))∞
n=0 = W(1, x, 2x, 1 + x ), (vn (x))∞
n=0 = W(0, 1, 2x, 1 + x ). (1)

Proof. The matrix  

x 1
M=
−1 x
has characteristic polynomial t2 −2xt+x2 +1. Consequently, it is immediate
to see that  
n un (x) vn (x)
M = .
−vn (x) un (x)
Using the matrix M we can observe that
    
un−1 (x) vn−1 (x) x 1 un (x) vn (x)
= ,
−vn−1 (x) un−1 (x) −1 x −vn (x) un (x)

i.e., (
un (x) = xun−1 (x) − vn−1 (x)
, ∀n ≥ 1.
vn (x) = un−1 (x) + xvn−1 (x)

2
Now, we prove the theorem by induction. It is straightforward to check that
!⊙2
1
1 v1 (x) 1 +1 2x v2 (x)
= , = x 1x = 2 = .
x u1 (x) x 1 − x2 x − 1 u2 (x)

Moreover, let us suppose

!⊙(n−1)
1 vn−1 (x)
=
x un−1 (x)

for a given integer n ≥ 1, then

!⊙n !⊙(n−1)
1 1 1 1 vn−1 (x) un−1 (x) + xvn−1 (x) vn (x)
= ⊙ = ⊙ = = .
x x x x un−1 (x) xun−1 (x) − vn−1 (x) un (x)

Theorem 2. Given n ∈ N and x ∈ R, with x 6= ±1, we have

!(−1)n
v n (x)
x⊙n = (−1)n+1 , ∀n ≥ 1
un (x)

where un (x) and vn (x) are given by Eq.(1).

Proof. By using the same arguments of Theorem 1, we can write

u1 (x) 2x v2 (x)
x= and x⊙2 = 1 =− .
v1 (x) 1− x2
u2 (x)
!(−1)n−1
vn−1 (x)
Let us suppose by induction that x⊙(n−1) = (−1)n , then
un−1 (x)
if n is even
vn−1 (x)
x−
un−1 (x) xun−1 (x) − vn−1 (x) un (x)
x⊙n = = = ,
vn−1 (x) un−1 (x) + xvn−1 (x) vn (x)
1+x
un−1 (x)
if n is odd
un−1 (x)
x+
vn−1 (x) xvn−1 (x) + un−1 (x) vn (x)
x⊙n = = =− .
un−1 (x) vn−1 (x) − xun−1 (x) un (x)
1−x
vn−1 (x)

3
 Let us highlight the matrix representation of the sequences (un )∞
n=0 and
(vn )∞
n=0 used in the previous theorem. Given the matrix
 
x 1
M=
−1 x

we have  
n un (x) vn (x)
M =
−vn (x) un (x)
   
n vm (x) vn+m (x)
M =
um (x) un+m (x)
The sequences (un )∞ ∞
n=0 and (vn )n=0 are particular cases of the Rédei poly-
nomials Nn (d, z) and Dn (d, z), introduced by Rédei [10] from the expansion
√ √ Nn (d, z)
of (z + d)n = Nn (d, z) + Dn (d, z) d. The rational functions have
Dn (d, z)
many interesting properties, e.g. , they are permutations of finite fields, as
described in the book of Lidl [7]. In [1], the authors showed that Rédei
polynomials are linear recurrent sequences of degree 2:
2 2
(Nn (d, z))∞
n=0 = W(1, z, 2z, z − d), (Dn (d, z))∞
n=0 = W(0, 1, 2z, z − d).

Thus, we can observe that

un (x) = Nn (−1, x), vn (x) = Dn (−1, x), ∀n ≥ 0.

Moreover, a closed expression of Rédei polynomials is well–known (see, e.g.,

[1]). In this way, we can derive a closed expression for the sequences (un )∞
n=0
and (vn )∞n=0 :

[n/2]  
 X n
(−1)k xn−2k


 u n (x) =
2k


k=0 . (2)
[n/2]  
 X n
(−1)k xn−2k−1

vn (x) =


2k + 1

k=0

Rational powers with respect to the product ⊙ can also be considered

by defining the n–th root as usual by

1
⊙
z=x n iff z ⊙n = x. (3)

Moreover, by means of Theorem 2, we have that Eqs. (3) are equivalent

to !(−1)n
n+1 vn (z)
x = (−1) ,
un (z)

4
i.e., by Eqs. (2), the n–th root of x with respect to the product ⊙ is a root
of the polynomial
n  
n k+1 1+(−1) k+1
(−1)⌊ 2 ⌋ x
X
Pn (z) = 2 zk .
k
k=0

Let us consider the equation

! !
1 1 π
nA +A = , (4)
x y 4

we want to solve it when n and x are integer values. We point out that Eq.
(4) is equivalent to
!⊙n
1 1
⊙ =1 (5)
x y
By Theorem 1 we have
!⊙n
1 1 vn (x) 1 un (x) + vn (x)y
⊙ = ⊙ = .
x y un (x) y −vn (x) + un (x)y

Thus
un (x) + vn (x)
y=
un (x) − vn (x)
solves Eq. (5), i.e.,
!⊙n
1 un (x) + vn (x)
⊙ = 1, ∀x ∈ Z
x un (x) − vn (x)

and consequently we can solve Eq. (4), i.e.,

! !
1 un (x) − vn (x) π
nA +A = + k(n, x)π, ∀x ∈ Z, (6)
x un (x) + vn (x) 4

where k is a certain integer number depending on n and x. Precisely, we

have ! !
1 π  
k(n, x) = sign nA − ⌊T ⌋ + χ( 1 ,1) ({T }) , (7)
x 4 2

where χ( 1 ,1) is the characteristic function of the set 21 , 1 and

2

!
π 1
− nA
4 x
T = .
π

5
In order to obtain Eq. (7), we can rewrite Eq. (6) as
! !
un (x) − vn (x) π 1
A = − nA + k(n, x)π.
un (x) + vn (x) 4 x

Let us consider the case! in which the first member lies in the interval
 π π π 1 π
− , . If − nA ≥ 0, then k(n, x) must be negative so that −
2 2 4 x 4
!
1
nA + k(n, x)π lies in the correct interval. Since
x
!
π 1
− nA = π (⌊T ⌋ + {T }) ,
4 x

1 π
it follows that if 0 ≤ {T } ≤ , then 0 ≤ π · {T } ≤ and consequently k =
2 2
1 π
−⌊T ⌋. Conversely, if < {T } < 1, then < π · {T } < π and, observing
2 2
that !
π 1
− nA = π (⌊T ⌋ + 1) + π ({T } − 1) ,
4 x
π
we obtain − < π({T } − 1) < 0, that is k(n, x) = −(⌊T ⌋ + 1).
2 !
π 1
Similar considerations apply to − nA < 0, obtaining Eq. (7).
4 x

Proposition 1. The sequences (un (x) + vn (x))∞ ∞

n=0 and (un (x) − vn (x))n=0
are linear recurrent sequences of order 2 and precisely
2 2
(un (x)+vn (x))∞
n=0 = W(1, x+1, 2x, 1+x ), (un (x)−vn (x))∞
n=0 = W(1, x−1, 2x, 1+x )

Proof. It immediately follows from the definition of the sequences (un )∞

n=0
and (vn )∞
n=0 .

Eq. (6) provides infinitely many identities that express π as sum of

arctangents.
Example 1. Taking n = 7 and x = 3 in Eq. (6) we have
! !
1 u7 (3) − v7 (3) π
7A +A = ,
3 u7 (3) + v7 (3) 4

i.e., ! !
1 278 π
7 arctan − arctan = .
3 29 4

6
For n = 8 and x = 3, we have
! !
1 863 π
8 arctan + arctan = + π.
3 191 4

For n = 5 and x = 2, we have

! !
1 79 π
5 arctan − arctan = .
2 3 4

For n = 2 and x = 7, we have

! !
1 17 π
2 arctan + arctan = .
7 31 4

2 Golden mean and π

In Mathematics the most famous numbers are π and the Golden mean.
Thus, it is very interesting to find identities involving these special numbers.
In particular, many expressions for π in terms of the Golden mean have
been found. For example, using the Machin formula of π via arctangents,
the following equalities arise
! !
π 1 1
= arctan + arctan
4 φ φ3
! !
π 1 1
= 2 arctan + arctan
4 φ2 φ6
! !
π 1 1
= 3 arctan + arctan
4 φ3 φ5
! !
1 1
π = 12 arctan + 4 arctan ,
φ3 φ5
see [3], [4], [5]. Moreover, in [8], the authors found all possible relations of
the form
π
= a arctan(φk ) + b arctan(φl ),
4
where a, b are rational numbers and k, l integers.
In this section, we find new expressions of π as sum of arctangents in-
volving φ. When n = 2, from Eq. (5) we find

x2 + 2x − 1
y= . (8)
x2 − 2x − 1

7
It is well–known that the minimal polynomial of φm is
fm (t) = t2 − Lm t + (−1)m ,
where (Lm )∞
m=0 = W(2, 1, 1, −1) is the sequence of Lucas numbers (A000032
in OEIS [11]). If we set x = φm in (8), then it is equivalent to replace
x2 + 2x − 1 and x2 − 2x − 1 with
x2 + 2x − 1 (mod fm (x)), x2 − 2x − 1 (mod fm (x)),
respectively. When m is odd, dividing by x2 − Lm x − 1, we obtain
(Lm + 2)x Lm + 2
y= =
(Lm − 2)x Lm − 2
and when m is even, we have
− 2 + (2 + Lm )x
y= ,
−2 + (−2 + Lm )x
and therefore
− 2 + (2 + Lm )φm
y= .
−2 + (−2 + Lm )φm
We find the following identities
! !
π 1 L2k+1 − 2
= 2 arctan + arctan (9)
4 φ2k+1 L2k+1 + 2
! !
π 1 − 2 + (L2k − 2)φ2k
= 2 arctan + arctan .
4 φ2k −2 + (L2k + 2)φ2k
The above procedure can be reproduced for any root α of a polynomial
x2 − hx + k, finding expression of π as the sum of arctangents involving
quadratic irrationalities.
√
Example 2. Let us express π in terms of 2. Its minimal polynomial is
x2 − 2 and
x2 + 2x − 1 (mod x2 − 2) = 1 + 2x, x2 − 2x − 1 (mod x2 − 2) = 1 − 2x.
We have ! √ !
π 1 1−2 2
= 2 arctan √ + arctan √ .
4 2 1+2 2
√
In general, if k is odd the minimal polynomial of 2k is x2 − 2k and
x2 +2x−1 (mod x2 −2k ) = 2k −1+2x, x2 −2x−1 (mod x2 −2k ) = 2k −1−2x.
We have the following identity
! k !
π 1 2k − 1 − 2 2 +1
= 2 arctan √ + arctan k
.
4 2k 2k − 1 + 2 2 +1

8
 1 √
Example 3. Let us consider α = (5 + 29). The minimal polynomial of
2
α3 is x2 − 140x − 1 and

x2 +2x−1 (mod x2 −140x−1) = 142x, x2 −2x−1 (mod x2 −140x−1) = 138x.

Thus, we have
! !
π 8 69
= 2 arctan √ + arctan .
4 (5 + 29)3 71

We can find different identities involving π and the Golden mean con-
sidering the equation
1
x⊙ 2 ⊙ y = 1. (10)

Proposition 2. For any real number x, the following equalities hold

p π
2A(−x ± 1 + x2 ) + A(x) = ± . (11)
2
Proof. By Theorem 2 we know that the roots of the polynomial P2 (z) =
1
xz 2 + 2z − x are the values of x⊙ 2 . Hence, from Eq. (10) we obtain

zi ⊙ y = 1, i = 1, 2, (12)

where √ √
−1+ 1 + x2 −1− 1 + x2
z1 = and z2 = .
x x
Finally, solving Eq. (10) with respect to y we get
p p
y1 = −x + 1 + x2 or y2 = −x − 1 + x2 .

It should be noted that if x is positive then y2 < 0 and z2 · y2 > 1 so that

1  1  π
A(x) + A(y2 ) = A x⊙ 2 + y2 − ,
2 2
similar reasoning can be applied if x is negative.
Now, substituting in Eqs. (12) we have

1 p π
A(x) + A(−x ± 1 + x2 ) = ± ,
2 4
or equivalently
p π
2A(−x ± 1 + x2 ) + A(x) = ± .
2

9
 Eqs. (11) yield to other interesting formulas involving π, φ and Lucas
numbers. To show this, we need some identities about Lucas numbers,
Fibonacci numbers and the Golden mean:
√
m Lm + Fm 5
φ = , L2m − 5Fm2
= 4(−1)m ,
2
see, e.g., [9]. Considering m odd, if we set
Lm
x=
2
it follows
p √
p − Lm − 4 + L2m − Lm − Fm 5
−x− 1+ x2 = = = −φm . (13)
2 2
Thus, substituting Eq. (13) into Eqs. (11) we find the formula
!
π L2k+1  
− = arctan − 2 arctan φ2k+1 . (14)
2 2
√
On the other hand, if we consider y = −x + 1 + x2 we have
p √
− L + 4 + L 2 − L + F 5
m m m m
p
− x + 1 + x2 = = . (15)
2 2
Moreover, √
− L m + F m 5 − L2m + 5Fm 2
φm · = = 1,
2 4
and substituting in Eqs. (11) another interesting formula arises
! !
π L2k+1 1
= arctan + 2 arctan . (16)
2 2 φ2k+1

Furthermore, by Eq. (9) we obtain an identity that only involves the Lucas
numbers ! !
π L2k+1 L2k+1 − 2
= arctan − arctan . (17)
4 2 L2k+1 + 2
The previous identity corresponds to a special case of the following propo-
sition.
Proposition 3. Let f, g be real functions. If
f (x) − 1
g(x) = ,
f (x) + 1
then
π
A(f (x)) − A(g(x)) = + kπ, (18)
4
for some integer k.

10
 π
Proof. We use the product ⊙ for solving A(f (x)) − A(g(x)) = . We have
4
!
f (x) − g(x) π
A =
1 + f (x)g(x) 4
and
f (x) − g(x)
=1
1 + f (x)g(x)
from which
f (x) − 1
g(x) = .
f (x) + 1

Remark 2. Eq. (18) has been found by means of only elementary algebraic
considerations. The same result could be derived from analysis. Observe
that given the functions f and g satisfying the hypothesis of the previous
proposition, then (arctan f (x))′ = (arctan g(x))′ .
When f (x) and g(x) are specified in Eq. (18), the value of k can be
retrieved as in Eq. (7) with analogous considerations.
The previous proposition allows to determine new beautiful identities.
ax ax − b
For example, the function f (x) = determines the function g(x) =
b ax + b
and ! !
ax ax − b π
A −A = + kπ.
b ax + b 4
For a = 1 and b = 2, we obtain the following interesting formulas
! !
π x x−2
= arctan − arctan , (19)
4 2 x+2
which holds for any real number x > −2 and
! !
3π x x−2
− = arctan − arctan , (20)
4 2 x+2
valid for any real number x < −2. Eqs. (19) and (20) provide infinitely
many interesting identities, like Eq. (17) and, e.g., the following ones
! !
π φ φ−2
= arctan − arctan
4 2 φ+2
! !
π Fm Fm − 2
= arctan − arctan
4 2 Fm + 2
√ ! √ !
π 2 2−2
= arctan − arctan √ .
4 2 2+2

11
References
[1] S. Barbero, U. Cerruti and N. Murru, Solving the Pell equation via Rédei
rational functions, The Fibonacci Quarterly 48 (2010) 348–357.

[2] J. S. Calcut, Gaussian integers and arctangent identities for π, The

American Mathematical Monthly 116(6) (2009) 515–530.

[3] H. C. Chan and S. Ebbing, π in terms of φ: Some recent developments,

Proc. of the Twelfth International Conference in Fibonacci Numbers and
Their Applications (San Francisco State University, 2006).

[4] H. C. Chan, π in terms of φ, The Fibonacci Quarterly 44(2) (2006)

141–145.

[5] H. C. Chan, Machin–type formulas expressing π in terms of φ, The

Fibonacci Quarterly 46/48(1), (2008/2009) 32–37.

[6] H. Chien–Lih, Some observations on the method of arctangents for the

calculation of π, The Mathematical Gazette 88(512) (2004) 270–278.

[7] R. Lidl, G. L. Mullen and G. Turnwald, Dickson polynomials, Pitman

Monogr. Surveys Pure appl. Math. 65 (Longman, 1993).

[8] F. Luca and P. Stanica, On Machin’s formula with powers of the Golden
section, International Journal of Number Theory 05(973) (2009).

[9] S. Rabinowitz, Algorithmic manipulation of Fibonacci identities, Appli-

cations of Fibonacci Numbers 6 (1996) 389–408.

[10] L. Rédei, Uber eindeuting umkehrbare polynome in endlichen korpen,

Acta Sci. Math. (Szeged) 11 (1946) 85–92.

[11] N. J. A. Sloane, The On–Line Encyclopedia of Integer Sequences,

Published electronically at http://www.research.att.com/njas/sequences
(2010).

[12] I. Tweddle, John Machin and Robert Simson on inverse–tangent series

for π, Arch. Hist. Exact Sci. 42 (1991) 1–14.

[13] J. W. Wrench, The evolution of extended decimal approximations of π,

Math. Teacher (1960) 644–650.

12