Chapter 2 Applications of Fourier series

One of the applications of Fourier series is the evaluation of certain infinite sums. For
P∞ 1 P∞ 1
example, 2
, 4
, are computed in Chapter 1 (see for example, Remark 2.4.1).
n=1 n n=1 n
In this chapter, we study several other applications of Fourier series.

2.1 The values of Riemann zeta function at even

positive integers
P∞ 1
The Riemann zeta function is defined by ζ(s) = s
, Re(s) > 1. It is
n=1 n
analytic in the region {s ∈ C : Re(s) > 1}.

Figure 2.1: The Riemann Zeta function.

One of the applications of Fourier series is evaluation of the values of Riemann

zeta function at even positive integers. In order to study this application, we make

69
use of Bernoulli numbers and polynomials. These Bernoulli numbers Bn are defined
1 n+1

as follows: We set B0 = 1 and for n ≥ 1, define Bn = − Bk . Bernoulli
n+1 k
polynomials are defined recursively as follows:

B0 (x) = 1,
0
Bn (x) = nBn−1 (x) for n ≥ 1,

and
Z1
Bn (x) dx = 0.
0

0
Let us compute B1 (x). By definition, B1 (x) = 1. On integration, we get B1 (x) =
R1
x + c, where c is a constant. But B1 (x) dx = 0. Hence, c = − 12 . In other words,
0
B1 (x) = x− 12 . Similarly Bn (x) can be calculated for n ≥ 2. The Bernoulli polynomials
satisfy the equation Bn (0) = Bn (1) = Bn ∀ n ≥ 2.

Functions on [0,1] can be considered as 1-periodic functions, i.e functions sat-

isfying f (x + 1) = f (x), on R and they can be expanded in terms of e2π ikx , k ∈ Z.
(See, for example, Remark 2.4.1.) Thus Bernoulli polynomials have a well defined
Fourier series.

Z1
B
b1 (k) = (x − 12 )e−2πikx dx
0
1
=− k 6= 0,
2πik
and
Z1
B
b1 (0) = (x − 21 ) dx = 0.
0

70
Thus the Fourier series of B1 (x) can be written as
X −1
B1 (x) = e2πikx .
k6=0
2πik

For n ≥ 2, the Fourier coefficients of Bn (x) are calculated as follows:

Z1
B
bn (k) = Bn (x)e−2πikx dx
0
1 Z1
Bn (x)e−2πikx  1
= + Bn0 (x)e−2πikx dx, k 6= 0
−2πik 0 2πik
0
Z1
1
= nBn−1 (x)e−2πikx dx
2πik
0
n b
= Bn−1 (k) k 6= 0,
2πik

and B
bn (0) = 0. Thus by induction we have

−n! X e2πikx
Bn (x) = .
(2πi)n k6=0 k n

For even integers 2n the above takes the form:

(2n)! X e2πikx
B2n (x) = −
(2π)2n (i)2n k6=0 k 2n
( −1 ∞
)
(−1)n+1 (2n)! X e2πikx X e2πikx
= +
(2π)2n k=−∞
k 2n
k=1
k 2n
( ∞ ∞
)
(−1)n+1 (2n)! X e−2πikx X e2πikx
= + .
(2π)2n k=1
k 2n
k=1
k 2n

71
Thus ( ∞
)
(−1)n+1 (2n)! X 1
B2n (x) = · 2cos(2πkx) .
(2π)2n k=1
k 2n

Evaluating at x = 0, we get

∞
(−1)n+1 2(2n)! X 1
B2n (0) = .
(2π)2n k=1
k 2n

Thus for even positive integers, m = 2n, n ∈ N, we get

(−1)n+1 (2π)2n
ζ(2n) = B2n · .
2(2n)!

2.2 Isoperimetric inequality

Among all the regions in the plane enclosed by a piecewise C 1 curve with a
given perimeter L the disc has the maximum area. If A is the area of the region then
we have 4πA ≤ L2 and this inequality is called the isoperimetric inequality. Equality
holds if and only if the curve is a circle.
In order to prove isometric inequality, first we state and prove Wirtinger’s
inequality.
Theorem 2.2.1. (Wirtinger’s inequality) Let f be a continuous and piecewise C1
function on T . Then

Zπ Zπ
2
|f (x) − fb(0)| dx ≤ |f 0 (x)|2 dx.
−π −π

Equality holds if and only if f (x) = fb(0) + fb(1)eix + fb(−1)e−ix .

72
Proof. Given that f is a continuous and piecewise C1 function on T . We can write
the Fourier series of f as

X
f (x) = fb(0) + fb(n)einx .
n6=0

From Plancheral theorem (Parseval identity) applied to fb(x) − fb(0) we get

Zπ 
1 2 X  2
f (x) − f (0) dx = f (n) .
 b  b 
2π
  
−π n6=0

Notice that f 0 (x) ∼ infb(n) einx . Since f 0 is piecewise continuous, f 0 is Riemann

P
n6=0
integrable on T . Hence, (f 0 )2 is Riemann integrable on T . This shows that f 0 ∈ L2 (T ).
Again applying Parseval identity, we get

Zπ
1 X
|f 0 (x)|2 dx = n2 |fb(n)|2 . (2.1)
2π n6=0
−π
Now,
Zπ Zπ
1 0 2 1
|f (x)| dx − |f (x) − fb(0)|2 dx
2π 2π
−π −π
X X
= n2 |fb(n)|2 − |fb(n)|2 ,
n6=0 n6=0
X
= (n2 − 1)|fb(n)|2 ≥ 0.
n6=0

This proves Wirtinger’s inequality. Clearly, equality holds if and only if (n2 − 1)
|fb(n)|2 = 0 for all n 6= 0. This means that fb(n) = 0 for all integers |n| ≥ 2. Thus
f (x) = fb(0) + fb(1) eix + fb(−1) e−ix .

73
 Now, we shall prove the isoperimetric inequality.
Suppose that the boundary curve Γ having length L is parameterized by the arc
length. In other words, the boundary curve Γ : z(s) = x(s) + i y(s) satisfies the
equation
 2  2
dx dy
+ = 1.
ds ds

Let

   
Lθ Lθ
f (θ) = x , g(θ) = y .
2π 2π

Then f and g are 2π-periodic functions on R. Further,

2   2
0 2 0 L 0 2 L 0
f (θ) + g (θ) = x (θ) + y (θ)
2π 2π
L2
= 2 x0 (θ)2 + y 0 (θ)2


4π
L2
= 2.
4π
Also note that
Zπ
g 0 (θ) dθ = g(π) − g(−π) = 0.
−π

Now, let Ω be the region enclosed by Γ. Then

Zπ Zπ
0
A = Area (Ω) = f (θ)g (θ) dθ = (f (θ) − fb(0))g 0 (θ) dθ
−π −π
 π 
Z Zπ
1 
≤ (f (θ) − fb(0))2 dθ + g 0 (θ)2 dθ
2 
−π −π

74
  π 
Z Zπ
1 
≤ f 0 (θ)2 dθ + g 0 (θ)2 dθ ,
2 
−π −π

using “ab ≤ 21 (a2 + b2 )” followed by Wirtinger’s inequality. Thus

Zπ
1
f 0 (θ)2 + g 0 (θ)2 dθ


A≤
2
−π
L2
= ,
4π

proving our assertion. Equality in isoperimetric inequality forces equality in Wirtinger’s

Rπ
inequality and equality in “ab ≤ 21 (a2 + b2 )” which yields (f − fb(0) − g 0 )2 dθ = 0.
−π
This implies that f (θ) = fb(0) + fb(1) eiθ + fb(−1) e−iθ and f − fb(0) − g 0 = 0.
Thus it follows that
fb(1) eiθ fb(−1) e−iθ
g(θ) = + + c,
i (−i)

where c is a constant of integration. Put fb(n) = α(n) + iβ(n). Then

Zπ Zπ
α(1) = Re f (θ) e−iθ dθ = f (θ) cos θ dθ = α(−1),
−π −π

and
Zπ Zπ
β(1) = Im f (θ) e−iθ dθ = f (θ) sin θ dθ = −β(−1).
−π −π

Thus f = fb(0) + 2α(1) cos θ − 2β(1) sin θ, and g = c + 2β(1) cos θ + 2α(1) sin θ. Thus
 2
L2
f − fb(0) + (g − c)2 = 4[α(1)2 + β(1)2 ]. On the other hand, f 0 (θ)2 + g 0 (θ)2 = 4π 2,

75
 L2 L2
which leads to 4[α(1)2 + β(1)2 ] = 4π 2
. Thus f − fb(0))2 + (g − c)2 = 4π 2
, showing that
the resulting curve is a circle.

2.3 Jacobi identity for the theta function

∞
exp (−πn2 t), t > 0. The Jacobi
P
The theta function is defined as θ(t) =
n=−∞
1
identity states that θ(t) = t− 2 θ 1t . In order to prove this identity, consider f (x) =


∞  
P −(x−k)2
exp 2t
. The series converges uniformly on [0, 1] and defines a 1-periodic
n=−∞
function on R, ı.e f (x + 1) = f (x) for all x ∈ R. We compute the Fourier coefficients
of f on the interval [0, 1]. We have

Z1 X
∞
−(x − k)2
 
fb(n) = exp e−2πinx dx
k=−∞
2t
0
∞ Z1
−(x − k)2
X  
= exp e−2πinx dx
k=−∞ 0
2t

∞ −k+1
−x2
X Z  
= exp e−2πinx dx
k=−∞ −k
2t
Z∞
−x2
 
= exp e−2πinx dx
2t
−∞
√ 2 n2 t
= 2πt e−2π .

The last inequality will be proved in Proposition 3.3.2 later. Clearly, fb ∈ l1 (Z). Thus

∞
√ X
exp −2π 2 n2 t e2πinx .


f (x) = 2πt
n=−∞

76
 t
Letting x = 0, and changing t to 2π
, we get

∞ ∞
√ X −πn2
X  
2
t exp (−πn t) = exp .
n=−∞ n=−∞
t
In other words,
1
θ(t) = t− 2 θ 1


t
.

2.4 Weierstrass approximation theorem

Before stating this theorem, we observe the following:
Remark 2.4.1. Let {kn } be a sequence of functions defined on [−1, 1] such that
R1
(i) kn (x) dx = 1 ∀ n ≥ 1.
−1
R1
(ii) |kn (x)| dx ≤ M ∀ n ≥ 1 for some M > 0.
−1 R
(iii) For every δ > 0, |kn (x)| dx → 0 as n → ∞.
δ≤|x|≤1
R1
Let f ∈ C[−1, 1]. Then f (x − t)kn (t) dt converges to f uniformly on [−1, 1]. The
−1
proof of this theorem is exactly the same as in Theorem 1.5.1.
Theorem 2.4.2. (Weierstrass approximation) Let f ∈ C[a, b]. Then there exists
a sequence of polynomials {pn } such that pn converges to f uniformly on [a, b].
R1
Proof. Let kn (x) = cn (1−x2 )n , x ∈ [−1, 1], where cn is chosen so that kn (x) dx = 1.
−1
Thus

Z1
1 = cn (1 − x2 )n dx
−1
Z1
= 2cn (1 − x2 )n dx
0

77
 √1
Zn
≥ 2cn (1 − x2 )n dx
0
√1
Zn
≥ 2cn (1 − nx2 ) dx
0
4cn
= √ ,
3 n

3√ √ √
which shows that cn ≤ 4
n < n. Fix δ > 0. Then kn (x) ≤ n(1 − δ 2 )n for
δ ≤ |x| ≤ 1, from which it follows that kn → 0 uniformly in δ ≤ |x| ≤ 1. In order to
prove the theorem, it is enough to assume that [a, b] = [0, 1], as there is a bijection
x−a
x 7→ b−a
between [a, b] and [0, 1]. Further it is enough to prove the theorem for such
function g for which g(0) = g(1) = 0. Assume for a moment that this theorem has
been proved for such functions.

Consider g(x) = f (x) − f (0) − x[f (1) − f (0)], x ∈ [0, 1]. Clearly, g(0) =
g(1) = 0. Thus g can be approximated by a sequence of polynomials which in turn
will approximate f . Thus, we assume that [a, b] = [0, 1] and f (0) = f (1) = 0. We
also define f (x) = 0 for x ∈
/ [0, 1]. Then f is uniformly continuous on R. Define

Z1
Pn (x) = f (x + t)kn (t) dt, x ∈ [0, 1]
−1
Z1−x
= f (x + t)kn (t) dt
−x
Z1
= f (t)kn (t − x) dt
0

78
 Z1
= f (t)kn (x − t) dt,
−1

as kn (−x) = kn (x). This shows that Pn (x) is indeed a polynomial. By the observation
made above, we conclude that Pn converges to f uniformly on [0, 1]. This proves the
theorem.

2.5 Wallis’ product formula

π
It is well known that Wallis’ product formula for 2
can be written as

π 2 2 4 4 6 6 8 8
= ···
2 1 3 3 5 5 7 7 9

This can be obtained by calculating the Fourier series of the functionf (x) = cos px
on [−π, π], where p is a real number but not an integer. Thus

Zπ
1
fb(n) = cos px e−inx dx
2π
−π
 π 
Z Zπ
1 
= ei(p−n)x dx + e−i(p+n)x dx
4π
−π −π
n
(−1) p
= sin pπ.
π p − n2
2

As fb ∈ l1 (Z) it is clear that the Fourier series of f converges pointwise [−π, π]. Thus

∞
p X (−1)n inx
cos px = sin (pπ) e
π n=−∞
p2 − n2

79
 " ∞
#
p 1 X (−1)n inx
= sin (pπ) 2 + 2 2 − n2
e
π p n=1
p
Letting x = π, we get,
" ∞
#
p 1 X 1
cos (pπ) = sin (pπ) 2 + 2 .
π p n=1
p2 − n2
In other words,
∞
X 2p 1
= π cot (pπ) − .
n=1
p2 −n 2 p

Integrating with respect to p, we get,

∞
p2
   
X sin (pπ)
ln 1 − 2 = ln
n=1
n pπ
In other words, we get,
∞ 
p2
Y 
sin (pπ) = pπ 1− 2 .
n=1
n

Now put p = 12 . We get the Wallis’ product formula for π2 .

2.6 Weyl’s equidistribution theorem

A sequence of real numbers ξ1 , ξ2 , . . . , ξn , . . . in [0,1) is said to be equidis-
tributed if for every interval [a, b) ⊂ [0, 1),

#{1 ≤ n ≤ N : ξn ∈ [a, b)}

lim = b − a.
N →∞ n

Theorem 2.6.1. (Weyl equidistribution theorem) A sequence of real numbers

ξ1 , ξ2 , . . . , ξn , . . . in [0,1) is equidistributed if and only if for all integers k 6= 0, one
N
has lim N1 e2πikξn = 0.
P
N →∞ n=1

80
 In order to prove this theorem, first we shall prove the following lemma.
Lemma 2.6.2. A sequence of numbers {ξn } in [0, 1) is equidistributed if and only if

N Z 1
1 X
lim f (ξn ) = f (x) dx,
N →∞ N
n=1 0

for every f ∈ L1 ([0, 1]).

1
PN R1
Proof. First assume that lim f (ξn ) = f (x) dx. Let [a, b] be an interval in
N →∞ N n=1 0
[0,1]. Then

Z1 Zb
χ[a,b) (x) dx = dx = b − a.
0 a
Further
N
X
χ[a,b) (ξn ) = #{1 ≤ n ≤ N : ξn ∈ [a, b)}.
n=1
Thus
N Z 1
1 X
lim χ[a,b) (ξn ) = χ[a,b) (x) dx,
N →∞ N
n=1 0

this shows that the sequence {ξn } is equidistributed. Conversely, suppose the sequence
{ξn } is equidistributed. Then as shown above

N Z 1
1 X
lim χ[a,b) (ξn ) = χ[a,b) (x) dx.
N →∞ N
N =1 0

But any step function on [0,1] is a linear combination of functions of the form χ[a,b) .
Hence, the result follows for every step function. If f ∈ L1 ([0, 1]), then given  > 0

81
there exist step functions s and t such that s ≤ f ≤ t and

Z1
(t(x) − s(x)) dx < .
0
Since the lemma holds for s,
N Z 1 Z 1
1 X
lim s(ξn ) = s(x) dx ≤  + f (x) dx.
N →∞ N
n=1 0 0

This means that there exists an N0 ∈ N such that for all N ≥ N0 ,

N Z 1
1 X
t(ξn ) ≤ 2 + f (x) dx.
N n=1
0
But
N N Z 1
1 X 1 X
f (ξn ) ≤ t(ξn ) ≤ 2 + f (x) dx.
N n=1 N n=1
0

1 PN R1
Similarly, we can show that f (ξn ) ≥ f (x) dx − 2. Thus, for all N ≥ N0 , we
N n=1 0
get,  
 XN Z1 
1 

N f (ξn ) − f (x) dx < 2,
 n=1 
0

proving our assertion.

Proof of Theorem. Since f : [0, 1] → C defined by f (x) = e2πinx belongs to L1 ([0, 1]).
R1
Further e2πinx dx = 0 for n 6= 0. If {ξn } is equidistributed, then
0

N
1 X 2πikξn
lim e = 0, (2.2)
N →∞ N
n=1

82
for each non zero integer k, by using lemma. Conversely, suppose (2.2) holds for
k 6= 0. Since any trigonometric polynomial f is a linear combination of functions of
the form e2πikx , we can conclude by linearity that

N Z 1
1 X
lim f (ξn ) = f (x) dx.
N →∞ N
n=1 0

But since every continuous periodic function can be approximated by a sequence of

trigonometric polynomials, the result follows for such continuous periodic functions.
But since such functions form a dense subset of L([0, 1]), the result follows. These
details are left as an exercise to the reader.

2.7 Exercises
In the following problems, ]x[ denotes the fractional part of x ∈ R.
1
2.7.1. Show that { ]n 2 [ : n = 1, 2, 3, . . .} is equidistributed in [0, 1).
 √   
n
1+ 3
2.7.2. Verify whether 2
: n = 1, 2, 3, . . . is equidistributed in [0, 1] or
not.
R1
2.7.3. Let f be a continuous function on [0, 1]. Suppose f (x) xn dx = 0, n =
0
1, 2, 3, . . . . Using Weierstrass theorem, show that f is identically zero on [0,1].

83
84