Basel problem: a physicist’s solution

arXiv:1908.07518v1 [math.HO] 20 Aug 2019

Zurab K. Silagadze

To paraphrase renowned American physicist Richard Feynman’s quote “ev-

ery theoretical physicist who is any good knows six or seven different theoretical
representations for exactly the same physics” [1], every mathematician who is
any good knows a dozen of solutions of the Basel problem, which asks for an
∞
1
P
evaluation of the infinite series n2 . For example, Moreno in the arXiv version
n=1
of [2] gives more than 80 references related to various proofs of Euler’s famous
formula, and some new ones have appeared since then [3, 4, 5, 6, 7, 8, 9]. Even
Euler himself gave at least four proofs [10, 11, 12] that
∞
X 1 π2
ζ(2) ≡ 2
= . (1)
n=1
n 6

In [13] Wästlund reformulated the Basel problem in terms of a physical system

using the proportionality of the apparent brightness of a star to the inverse
square of its distance. Inspired by this approach, here we give another physical
interpretation which, in our opinion, is simpler, natural enough, and leads to a
proof of (1) which is very Eulerian in its spirit.

A physicist’s solution of the Basel problem

First of all, let’s notice [13] that, because of
∞ ∞ ∞
X 1 X 1 1X 1
= + ,
n=1
n2 n=1
(2n − 1)2 4 n=1 n2

we have ∞ ∞ ∞
X 1 4X 1 1X 1
= = . (2)
n=1
n 2 3 n=1 (2n − 1)2 3 n=1 (n − 21 )2

A physicist can interpret (2) as representing a Coulomb force exerted on a unit

charge located at a point x = 1/2 by a semi-infinite sequence of opposite charges
located at points x1 = 1, x2 = 2, . . . The corresponding electrostatic potential
is ∞
X 1
U (x) = − , (3)
n=1
n−x

1
and we can write
∞   
X 1 1  1 dU (x) 
= F (x)
 = −  1. (4)
n=1
n2 3 x= 1 3 dx x=
2 2

Unfortunately (3) diverges and hence we are returning here (temporarily) to the
standards of mathematical rigor of Eulerian times. However physicists are used
to infinities, and thus let’s regularize the potential (3):
∞  
X 1 1
U (x) → UR (x) = U (x) − U (0) = − − . (5)
n=1
n−x n

Note that our regularization procedure does not affect at all the force and hence
we have ∞  
d X 1 1
F (x) = − . (6)
dx n=1 n − x n
In (6) we recognize immediately the presence of the digamma function ψ(x) =
Γ′ (x)/Γ(x) because
∞  
X 1 1
ψ(1 − x) = −γ − − , (7)
n=1
n−x n

γ being the Euler constant (this relation will be discussed below). Therefore
d
F (x) = [−ψ(1 − x) − γ] = ψ1 (1 − x),
dx
where ψ1 (x) = ψ ′ (x) is the trigamma function. Fortunately (4) indicates that
we need the trigamma function at x = 1/2:
∞  
X 1 1 1
2
= ψ1 , (8)
n=1
n 3 2

and this is just the value at which the trigamma function can be simply calcu-
lated thanks to the Euler’s reflection formula:
π2
ψ1 (1 − x) + ψ1 (x) = , (9)
sin2 πx
1 π2


which gives ψ1 2 = 2 , and hence from (8) we immediately get Euler’s famous
formula (1).

Several remarks
Above we presented a physics-motivated approach to the Basel problem. Of
course the connection with physics is tenuous at best. However the interpre-
tation of inverse squares as representing Coulomb forces was a crucial insight

2
in defining a connection with polygamma functions and the reflection formula.
The resulting formalism is, in fact, quite elementary, in the sense that its basic
pillars (7) and (9) can be obtained by elementary means.
For example, (7) follows from Newman’s infinite product formula (used by
Weierstrass as his definition of the gamma function)
∞ 
1 Y z  −z/n
= eγz 1+ e , (10)
Γ(1 + z) n=1
n

which by itself is just another version of Euler’s definition of the gamma function
as the limit [14]
nz n!
Γ(z) = lim . (11)
n→∞ z(z + 1) · · · (z + n)

A simple consequence of (11) is the following interesting identity [15]

∞
Y (k + α1 ) · · · (k + αn ) Γ(β1 ) · · · Γ(βn )
= , (12)
(k + β1 ) · · · (k + βn ) Γ(α1 ) · · · Γ(αn )
k=0

where n ≥ 1 and α1 , . . . , αn , β1 , . . . , βn are nonzero complex numbers, none of

which are negative integers, such that α1 +. . .+αn = β1 +· · ·+βn . In particular,
when α1 = 1 + z, α2 = 1 − z, β1 = β2 = 1, we get
∞ 
z2

Y 1
1− 2 = ,
n=1
n Γ(1 + z) Γ(1 − z)

which, in combination with Γ(1 + z) = z Γ(z) and Euler’s celebrated formula

∞ 
z2

Y sin πz
1− 2 = , (13)
n=1
n πz

implies the validity of the reflection formula

π
Γ(z) Γ(1 − z) = , (14)
sin πz
from which other reflection formulas, like (9), do follow.
It is tempting to consider the infinite product formula (13) to be a real back-
bone of the presented approach, as in the Euler’s original first proof. Although
several elementary proofs of Euler’s infinite product for the sine exist in the
literature (see, for example, [16, 17, 18, 19]), they do not seem to be signifi-
cantly simpler than the original proof by Euler. Therefore it may appear that
the reflection formula is fairly nontrivial to derive and its proof is as hard a
problem as the one we seek to solve. However this is actually not the case. It
is possible to avoid the use of the Euler’s infinite product in the derivation of
the reflection formula. Below we provide one such proof which is simple and
elementary enough and doesn’t rely on the infinite product formula for the sine
function.

3
 This proof of the reflection formula was inspired by Richard Dedekind’s 1852
proof [20] of (14) which seems to be not as well known as it deserves to be. It
appears as an exercise in [21] and was popularized in [22]. We prove not (14),
but the reflection formula for the digamma function

ψ(x) − ψ(1 − x) = −π cot πx, (15)

from which (9) follows by differentiation.

During the proof, which seems to be much simpler than the Dedekind’s
original one, we freely interchange the order of integrals and differentiate under
the integral signs, as physicist are generally accustomed to doing. A genuine
mathematician, of course, will resort in these cases to Fubini’s theorem and to
Lebesgue’s dominated convergence theorem to justify these operations [22].
We have the following well known integral representation for the digamma
function: Z 1
1 − t−x
ψ(1 − x) = −γ + dt. (16)
0 1−t
Indeed, expanding 1/(1 − t) in geometric series, interchanging the order of sum-
mation and integration and thus integrating term by term, we get (7).
Therefore,
Z 1 −x
t − tx−1 1 ∞ t−x − tx−1
Z
φ(x) ≡ ψ(x) − ψ(1 − x) = dt = dt. (17)
0 1−t 2 0 1−t
The validity of the last step can be checked by breaking the corresponding
integral into two integrals, over (0, 1) and (1, ∞), and putting y = 1/t in the
second integral.
Naively, the two parts of the last integral in (17)
Z ∞ −x Z ∞ x−1
t −t
dt and dt, (18)
0 1 − t 0 1−t
appear to be the same, because the second transforms into the first under the
change of integration variable y = 1/t. However the separate integrals in (18) are
ill-defined because of a singularity at t = 1. Nevertheless these integrals become
well-defined and equal in the sense of Cauchy principal value. Therefore,
Z ∞ −x
t
φ(x) = P dt. (19)
0 1 −t
To get rid of inconvenient principal value, we use quantum physicists’ favorite
formula (Sokhotski-Plemelj formula1 [23, 24])
1 1
= P ∓ iπδ(z), (20)
z ± iǫ z
1 Sokhotski-Plemelj formula is a relation between the generalized functions, that is it is

assumed that both sides of (20) are multiplied by a smooth function, which is non-singular
in a neighborhood of the origin, then integrated over a range of z containing the origin, and
finally a limit ǫ → 0 is taken in the results.

4
which gives
∞
t−x
Z
φ(x) ± iπ = lim dt. (21)
ǫ→0 0 1 − t ∓ iǫ
Multiplying these two representations of φ(x), we get
Z ∞ Z ∞
2 2 t−x s−x
φ (x) + π = lim dt ds. (22)
ǫ→0 0 1 − t + iǫ 0 1 − s − iǫ
We can substitute s = y/t in the second integral, interchange the order of
integrations, solve the resulting simple integral in t,
Z ∞
1 ln y
lim dt = − , (23)
ǫ→0 0 (1 − t + iǫ)(t − y − iǫt) 1−y
and end up with
∞
y −x ln y
Z
2 2
φ (x) + π = − dy. (24)
0 1−y
On the other hand, if we differentiate (19) by x, we get
Z ∞ −x
′ dφ(x) t ln t
φ (x) ≡ =− dt. (25)
dx 0 1−t
(there is no longer a need for the principal value after differentiation, because
the singularity softens and becomes integrable). Comparing (24) and (25), we
see that the function φ(x) satisfies differential equation
φ′ (x) = π 2 + φ2 (x). (26)
Note that this differential equation is much simpler than the differential equation
F (x)F ′′ (x) = (F ′ (x))2 + F 4 (x) obtained by Dedekind in [20] for the function
F (x) = Γ(x)Γ(1 − x).
Under the initial condition φ(1/2) = 0, which follows from the definition of
φ(x), (26) can be solved immediately:
Z φ
1 dφ 1 φ
x− = 2 + φ2
= arctan . (27)
2 0 π π π
Therefore  π
φ(x) = π tan πx − = −π cot πx, (28)
2
and this completes the proof of (15).

Zeta function values at positive even integers

The above approach can be easily generalized (this time without any physics
input) to enable a calculation of all ζ(2k). Because of
∞ ∞ ∞
X 1 X 1 1 X 1
= + ,
n=1
n2k n=1
(2n − 1)2k 22k n=1 n2k

5
we have
∞ ∞ ∞
X 1 22k X 1 1 X 1
ζ(2k) ≡ = = . (29)
n=1
n 2k 2k
2 − 1 n=1 (2n − 1)2k 2 − 1 n=1 (n − 21 )2k
2k

On the other hand, differentiating (7) 2k − 1 times, we get

∞
X 1
ψ2k−1 (1 − x) = (2k − 1)! , (30)
n=1
(n − x)2k

dn
where ψn (x) = dxn ψ(x) is the polygamma function. Therefore

ψ2k−1 21


ζ(2k) = 2k . (31)
(2 − 1)(2k − 1)!

Since
π2 d
2 = −π cot πx,
sin πx dx
differentiating (9) 2k−) times, we get

d2k−1
ψ2k−1 (1 − x) + ψ2k−1 (x) = −π cot πx. (32)
dx2k−1
It follows from this reflection formula that
 
1 π
ψ2k−1 = − s2k−1 , (33)
2 2

where the sn numbers are defined through

dn


sn = cot πx . (34)
dxn x= 12

Thanks to identity tan π x − 12 = − cot πx, we can express the sn numbers

through more familiar tangent numbers [25]

dn


Tn = tan x (35)
dxn 
x=0

as sn = −π n Tn and (31) takes the form

π 2k T2k−1
ζ(2k) = . (36)
2(22k − 1)(2k − 1)!

We can calculate tangent numbers recursively. To do so, note that

dn dn−1 1 dn−1 dn−1

= n−1 1 + tan2 x = n−1 (tan x tan x) ,


n
tan x = n−1 2
dx dx cos x dx dx

6
and apply the Leibniz formula for n-th derivative of a product of two functions.
We get the recurrence relation
n−1
X 
n−1
Tn = Tr Tn−1−r . (37)
r=0
r

From the definition (35) we find T0 = 0, T1 = 1 and it is not hard to prove by

induction that (37) implies the vanishing of tangent numbers if their index is
even. For an odd index, let’s take r = 2m − 1 in (37) to transform it into the
form [25]
k−1
X  2k − 2 
T2k−1 = T2m−1 T2(k−m)−1 . (38)
m=1
2m − 1

In principle (36) and (38) solve the generalized Basel problem and allow to
calculate ζ(2k) at least for small values of k. For example, we easily get T3 =
2, T5 = 16, T7 = 272, T9 = 7936, T11 = 353792 which imply

π4 π6 π8 π 10 691π 12
ζ(4) = , ζ(6) = , ζ(8) = , ζ(10) = , ζ(12) = .
90 945 9450 93555 638512875
However for large values of k more efficient algorithms are needed to calculate
tangent numbers. One of them can be found in [26].
A connection with Bernoulli numbers is established by the well-known for-
mula (see, for example, [26]), valid for n > 1,

n Tn−1
Bn = − .
(2i)n (2n − 1)

We also can use (31), (33) and

(2π)2n−1 (22n − 1)
sn = (−1)n Bn ,
n
proved in [27]. In either way we get the well known result

(2π)2k B2k
ζ(2k) = (−1)k+1 .
2 (2k)!

Recurrence formula for ζ(2k)

With some extra effort, it is possible to obtain a nice recurrence relation for
ζ(2k) [28, 29] which allows a calculation of ζ(2k) recursively, and thus also
provides a solution of the generalized Basel problem.
Let’s introduce another sequence of numbers related to the cotangent func-
tion:
dn
Sn = n (x cot x)|x=0 . (39)
dx

7
In [28] the coefficients of the Taylor expansion of x cot x were related to the
values of ζ(2k). As an alternative to that argument, we will study the numbers
Sn and, using the insights from the previous section, relate them to ζ(2k).
Because of relation
x tan x = x cot x − 2x cot 2x, (40)
which can be simply checked, the numbers introduced are related to the tangent
numbers. Namely, differentiating (40) n times and setting x = 0 produces the
equality n Tn−1 = Sn − 2n Sn . Therefore
22k − 1
T2k−1 = − S2k
2k
and
π 2k S2k
ζ(2k) = − . (41)
2(2k)!
Now we obtain a recurrence relation for the Sn . By using cot′ x = −(1 + cot2 x)
it can be checked that [28]
x (x cot x)′ = x cot x − x2 cot2 x − x2 . (42)
Differentiating both sides of this relation n > 2 times and setting x = 0, we get
n   n−1
X n
X n
nSn = Sn − Sr Sn−r = Sn − 2Sn − Sr Sn−r .
r=0
r r=1
r

Therefore the Sn numbers obey the recurrence relation

n−1
X n
(n + 1)Sn = − Sr Sn−r . (43)
r=1
r

As S0 = 1, S1 = 0, it follows from this recurrence relation (by induction) that

Sn = 0, if n is odd (if n is odd, one of the numbers r, n − r is also odd).
Therefore, taking r = 2m, we can write the recurrence relation (43) in the form
k−1
X  2k 
(2k + 1)S2k = − S2m S2k−2m . (44)
m=1
2m

If we substitute (44) into (41), we get a recurrence relation for the zeta-function
that was called “highly elegant” in [29]:
  k−1
1 X
k+ ζ(2k) = ζ(2m) ζ(2k − 2m). (45)
2 m=1

In fact (45) is equivalent to Euler’s recurrence relation for Bernoulli numbers

(independently found by Ramanujan [30])
n−1
X  2n 
(2n + 1)B2n = − B2m B2n−2m . (46)
m=1
2m

8
Both (45) and (46) were rediscovered many times [31]. For example Williams
[32] thought he was the first to explicitly state the recurrence relation in the form
(45). Actually, this recurrence relation is given implicitly in Euler’s work [33]
(paper E130 at http://www.math.dartmouth.edu/~euler/), and is explicitly
stated at least as early as 1906 in the book [34] (with the remark that this
recurrence relation is well known). Nowadays the proof of the recurrence relation
(45) is often given as an exercise in number theory courses (see, for example,
[35]).

Acknowledgments
The work is supported by the Ministry of Education and Science of the Russian
Federation. The author thanks Professor Juan Arias de Reyna for indicating
several interesting references, as well as an anonymous referee for constructive
comments which helped to improve the presentation.

References
[1] Feynman, R. (2017). The Character of Physical Law. Cambridge, MA: MIT
Press, p. 168.
[2] Moreno, S. (2016). A short and elementary proof of the Basel problem,
College Math. J. 47(2):134-135. https://arxiv.org/abs/1502.07667
[3] Boyadzhiev, K.N. (2015). Some integrals related to the Basel problem, Ci-
entia. A 26:1-13.
[4] Glebov, G. (2015). A peculiar proof of an identity of Euler. Math. Gaz.
99(544):139-143.
[5] Ghorbanpour, A., Hatzel, M. (2017). Parseval’s Identity and Values of Zeta
Function at Even Integers, https://arxiv.org/abs/1709.09326
[6] Costandin, M. (2017). A new proof for the exact values of ζ(2k) for k ∈ N,
https://arxiv.org/abs/1712.02255
R∞ sin x
[7] Siklos, S. (2018). A method of evaluating ζ(2) and x dx, Math. Gaz.
0
102(553):114-121.
[8] Kapil R. Shenvi Pause (2018). Basel Problem: A Solution Motivated by
the Power of a Point, Amer. Math. Monthly 125(6): 558-560.
[9] Vermeeren, M. (2018). Modified Equations and the Basel Problem. Math.
Intelligencer 40(2):33-37.

9
[10] Sandifer, C.E. (2007). Euler’s solution of the Basel problem — the longer
story, in Euler at 300: An Appreciation. Washington: Mathematical Asso-
ciation of America, pp. 105-117.
[11] Lagarias J.C. (2013). Euler’s constant: Euler’s work and modern develop-
ments, Bulletin Amer. Math. Soc. 50(4):527-628.
[12] Roy, R. (2011). Sources in the Development of Mathematics: Infinite Series
and Products from the Fifteenth to the Twenty-first Century. Cambridge:
Cambridge University Press.
[13] Wästlund, J. (2010). Summing Inverse Squares by Euclidean Geometry,
http://www.math.chalmers.se/~wastlund/Cosmic.pdf
[14] Jameson, G.J.O. (2014). A fresh look at Euler’s limit formula for the gamma
function, Math. Gaz. 98(542):235-242.
[15] Chamberland, M., Straub, A. (2013). On gamma quotients and infinite
products, Adv. in Appl. Math. 51:546-562.
[16] Eberlein, W.F. (1977). On Euler’s infinite product for the sine, J. Math.
Anal. Appl. 58(1):147-151.
[17] Feller, W. (1967). A Direct Proof of Stirling’s Formula, Amer. Math.
Monthly, 74(10):1223-1225.
[18] Ciaurri, Ó. (2015). Euler’s Product Expansion for the Sine: An Elementary
Proof, Amer. Math. Monthly 122(7):693-695.
[19] Salwinski, D. (2018). Euler’s Sine Product Formula: An Elementary Proof,
College Math. J. 49(2):126-135.
[20] Dedekind, R. (1852). Über ein Eulersches In-
tegral, J. Reine Angew. Math. 45:370-374.
https://archive.org/details/journalfrdierei96crelgoog/page/n385
[21] Andrews, G.E., Askey, R., Roy, R. (1999). Special Functions. Cambridge:
Cambridge University Press, pp. 49-50.
[22] Srinivasan, G.K. (2011). Dedekind’s Proof of Euler’s Reflection Formula
via ODEs, Mathematics Newsletter 21(3):82-83.
[23] Vladimirov, V.S. (1971). Equations of Mathematical Physics. New York:
Marcel Dekker, pp. 75-76.
[24] Weinberg, S. (1995). The Quantum Theory of Fields. Volume I: Founda-
tions. Cambridge: Cambridge University Press, pp. 112-113.
[25] Qi, F. (2015). Derivatives of tangent function and tangent numbers, Appl.
Math. Comput. 268:844-858.

10
[26] Knuth, D.E., Buckholtz, T.J. (1967). Computation of tangent, Euler, and
Bernoulli numbers, Math. Comp. 21: 663-688.
[27] Cvijović, D. (2009). Values of the derivatives of the cotangent at rational
multiples of π, Appl. Math. Lett. 22:217-220.
[28] Koecher, M. (1987). Klassische elementare Analysis. Basel:Birkhäuser, p.
157.
[29] Brede, M. (2007). Eulers Identitäten für die Werte von ζ(2n), Math.
Semesterber. 54: 135-140.
[30] Berndt, B.C. (1985). Ramanujan’s Notebooks: Part 1. New York: Springer-
Verlag, p. 122.
[31] Sitaramachandrarao, R., Davis, B. (1986). Some identities involving the
Riemann zeta function, II, Indian J. Pure Appl. Math. 17(10): 1175-1186.
[32] Williams, G.T. (1953). A New Method of Evaluating ζ(2n), Amer. Math.
Monthly 60(1):19-25.
[33] Lord, N. (2014). Beyond the Basel problem: Euler’s derivation of the gen-
eral formula for ζ(2n), Math. Gaz. 98(543):459-474.
[34] Nielsen N., (1906). Handbuch der theorie der gammafunktion. Leipzig:
Druck und Verlag von B.G. Teubner, p. 49.
[35] Ribenboim, P. (2001). Classical Theory of Algebraic Numbers. New York:
Springer, p. 503.

Zurab K. Silagadze
Budker Institute of Nuclear Physics and Novosibirsk State University, 630
090, Novosibirsk, Russia. [email protected]

11