You are on page 1of 5

Another Proof of ${\zeta(2)=\frac{\pi^2}{6}}$ Using Double Integrals

Author(s): Daniele Ritelli


Source: The American Mathematical Monthly, Vol. 120, No. 7 (AugustSeptember 2013),
pp. 642-645
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/10.4169/amer.math.monthly.120.07.642
Accessed: 19-04-2017 03:02 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access
to The American Mathematical Monthly

This content downloaded from 200.49.224.88 on Wed, 19 Apr 2017 03:02:09 UTC
All use subject to http://about.jstor.org/terms
NOTES
Edited by Sergei Tabachnikov

Another Proof of (2) = 6


2

Using Double Integrals


Daniele Ritelli

Abstract. Starting from the double integral


dx dy
Z Z
,
0 0 (1 + y)(1 + x 2 y)
we give another solution to the Basel Problem

X 1 2
(2) = = .
n=1
n2 6

The celebrated Euler identity, known as the Basel Problem,



X 1 2
= , (1)
n=1
n2 6

has been proved in many different ways. In this note, we focus on the derivation of
(1), taking advantage of the nice interplay between a double integral and a geometric
series, as has appeared in several articles on this subject [1, 5, 3]. As Sir Michael
Atiyah declared in an interview [8]:

Any good theorem should have several proofs, the more the better. For two rea-
sons: usually, different proofs have different strengths and weaknesses, and they
generalize in different directions: they are not just repetitions of each other. . . .

We find that there is always something worthy of attention in a new proof of a known
result. Here, we provide a proof that uses a rational function with the lowest degree
among the functions used in different proofs of the same kind.
The author who inaugurated this approach was Apostol [1], inspired by Beukers
paper [2], where the double integral
1 1
dx dy
Z Z
(2)
0 0 1 xy
was used to prove the irrationality of (2). Apostol, instead, evaluated (2) in two ways:
first, by expanding 1/(1 x y) into a geometric series, and then with a change of
variables corresponding to the rotation of the coordinate axes through the angle /4
http://dx.doi.org/10.4169/amer.math.monthly.120.07.642
MSC: Primary 40A25

642 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120


This content downloaded from 200.49.224.88 on Wed, 19 Apr 2017 03:02:09 UTC
All use subject to http://about.jstor.org/terms
radians. By equating the expression so obtained, the value of (2) is found. It is worth
noting that for the second evaluation, we have to compute the elaborate integrals

1/ 2
1 u
Z  
I1 = arctan du
0 2 u2 2 u2

and

2
!
1 2u
Z
I2 = arctan du,
1/ 2 2 u2 2 u2

using suitable trigonometric changes of variables. The evaluation of Beukers, Calabi,


and Kolk is similar [3]. They expand 1/(1 x 2 y 2 ) into a geometric series to obtain
the (2) series, after which they introduce the two-dimensional trigonometric changes
of variables
sin u sin v
x= , y= ,
cos v cos u
to evaluate the double integral. The proof of Hirschhon [6] stems from the double
inequality
2
a a 4n+2

< 2 (arctan a)2 , for 0 < a < 1, (3)
X
2 arctan <
1 + 1 a2 n=0
(2n + 1)2

and a passage to the limit as a 1. The derivation of (3) is based on an integral


inequality with regard again to the function f (x, y) = 1/(1 x 2 y 2 ). To obtain (3),
two integrals of f (x, y) over two different regions of the plane are computed. All
these approaches have in common the need to remove a singularity at the point (1, 1)
of the integrand. Our proof is inspired by [5], where another definite integral
1
+
x
Z Z
dx dy (4)
0 0 (1 + x 2 )(1 + x 2 y2)

is computed, first by integrating with respect to x and then with respect to y, and vice
versa. The same integral is considered in the probabilistic proof given in [7], where
integral (4) comes from the product of two positive Cauchy random variables. The
proof of [5], as well our proof, uses functions with no singularity in the domain of
integration, so we can consider that in same sense these proofs are simpler. Moreover,
our proof uses a lower-degree rational function than the one used in [5].
Our starting point, as with most of the papers on this subject, is that (1) is equiva-
lent to

X 1 2
= . (5)
n=0
(2n + 1)2 8

In our proof we will show (5) starting from the double integral

dx dy
Z Z
. (6)
0 0 (1 + y)(1 + x 2 y)

AugustSeptember 2013] NOTES 643

This content downloaded from 200.49.224.88 on Wed, 19 Apr 2017 03:02:09 UTC
All use subject to http://about.jstor.org/terms
If we integrate (6) first with respect to x and then to y, we find that


1
dx
1 arctan( y x) x=
Z  Z  Z    
dy = dy
0 1+y 0 1 + x2y 0 1+y y x=0


dy
Z
=
2 0 y(1 + y)

2u 2
Z
= du = , (7)
2 0 u(1 + u 2 ) 2

where we used the change of variable y = u 2 in the last step. Reversing the order of
integration yields
Z
dy 1 1 x2
Z  Z Z   
dx = dy dx
0 0 (1 + y)(1 + x 2 y) 0 1x
2
0 1+y 1 + x2y
1 1 ln x
Z Z
= ln dx = 2 dx. (8)
0 1 x 2 x 2
0 x 2 1

Hence, equating (7) and (8), we get



ln x 2
Z
dx = . (9)
0 x2 1 4

Now split the integration domain in (9) between [0, 1] and [1, ) and change the
variable x = 1/u in the second integral, so that
1

ln x ln x
ln x
Z Z Z
dx = dx + dx
0 x2 1 0 x2 1 1 x2 1
1 1
ln x ln u
Z Z
= dx + du. (10)
0 x 1
2
0 u2 1

From (9) and (10) we get


1
ln x 2
Z
dx = . (11)
0 x2 1 8

Equation (5) now follows, expanding, as in [5], the denominator of the integrand on
the left-hand side of (11) into a geometric series and using the Monotone Convergence
Theorem (see [4, pp. 9596]). Thus, we have:

1 1 + 1
ln x ln x
Z Z Z
(x 2n ln x) dx.
X
dx = dx = (12)
0 x2 1 0 1 x2 n=0 0

Integrating by parts yields

1 1 Z 1
x 2n+1 x 2n 1
Z 
2n
(x ln x) dx = ln x + dx = , (13)
0 2n + 1 0 0 2n + 1 (2n + 1)2

644 c THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 120


This content downloaded from 200.49.224.88 on Wed, 19 Apr 2017 03:02:09 UTC
All use subject to http://about.jstor.org/terms
so that considering (13), we can write (12) as

1 +
ln x 1
Z X
dx = , (14)
0 x 1
2
n=0
(2n + 1)2

and we are done equating (11) and (14).

REFERENCES

1. T. Apostol, A proof that Euler missed: Evaluating (2) the easy way, Math. Intelligencer 5 (1983) 5960,
available at http://dx.doi.org/10.1007/BF03026576.
2. F. Beukers, A note on the irrationality of (2) and (3), Bull. London Math. Soc. 11 (1979) 268272,
available at http://dx.doi.org/10.1112/blms/11.3.268.
3. F. Beukers, E. Calabi, J.A.C. Kolk, Sums of generalized harmonic series and volumes, Nieuw Arch. Wisk.
11 (1993) 217224.
4. M. Capinski, E. Kopp, Measure, Integral and Probability, second edition, Springer, New York, 2004.
2
5. J.D. Harper, Another simple proof of 1 + 212 + 312 + = 6 , Amer. Math. Monthly 110 (2003) 540541,
available at http://dx.doi.org/10.2307/3647912.
2
6. M.D. Hirschhon, A simple proof that (2) = 6 , Math. Intelligencer 22 (2011) 8182, available at http:
//dx.doi.org/10.1007/s00283-011-9217-4.
2
7. L. Pace, Probabilistically proving that (2) = 6 , Amer. Math. Monthly 118 (2011) 641643, available at
http://dx.doi.org/10.4169/amer.math.monthly.118.07.641.
8. M. Raussen, C. Skau, Interview with Michael Atiyah and Isadore Singer, Notices Amer. Math. Soc. 52
(2005) 225233.

Dipartimento di Scienze Statistiche, Universit`a di Bologna, Viale Filopanti, 5 Bologna, Italy


daniele.ritelli@unibo.it

Simple Approximation Formulas


for the Birthday Problem
Matthias Arnold and Werner Gla

Abstract. We approximate the probability for at least k = 2 or 3 of n persons having the same
birthday if we ignore leap years and assume that each day is equally likely. Although there are
exact formulas available in the literature, a possible drawback to using them in the classroom
is that these formulas are comparatively cumbersome. As an alternative, we suggest an easily
computable approximation that still provides a close fit. Furthermore, the approximation al-
lows an easy answer to the question of how many people are needed, so that the probability of
at least two or three of them having the same birthday exceeds a given value.

The famous birthday problem is concerned with the probability of matching birthdays
in a randomly selected group of people. Most students are surprised to learn that the
probability for at least one match exceeds 0.5 if more than 22 persons are involved.
Numerous authors generalize the birthday problem to, for example, more than two
http://dx.doi.org/10.4169/amer.math.monthly.120.07.645
MSC: Primary 62E17, Secondary 41A10

AugustSeptember 2013] NOTES 645

This content downloaded from 200.49.224.88 on Wed, 19 Apr 2017 03:02:09 UTC
All use subject to http://about.jstor.org/terms