Fourier Series

Grados en Ingenierías de Telecomunicacón

A lot of physical problems, in particular, problems concerning to the waves and signals use
of the so-called Fourier Series which we proceed to study. We will see as this idea inspires the
Fourier Transform.

1 Fourier Series for period 2π

A function f : R → R is periodic function with period T > 0 if f (x+nT ) = f (x) for all n integer.

x0 T x0 + T

Figure 1: Periodic function of period T .

These known definitions will be used:

Even function: If f (−x) = f (x) for all x ∈ R. Examples: x2 , cos x, …

Odd function: If f (−x) = −f (x) for all x ∈ R. Examples: x5 , sin x…

Expanding a function as a trigonometric series is sometimes more advantageous than ex-

panding it as a power series. In particular, astronomical phenomena are usually periodic, as
are electromagnetic waves, and vibrating strings, so it makes sense to express them in terms of
periodic functions.

Definition 1.1. Let f (x) a periodic function with period 2π, we say that f admits a trigonomet-
ric expansion in Fourier series if there exist sequences {an }, n = 0, 1, . . . and {bn }, n = 1, 2, . . . ,
called Fourier coefficients such that
∞
a0 X
f (x) = + ak cos (kx) + bk sin (kx) (1)
2
k=1

Assuming that the trigonometric series converges and has a continuous function as its sum on
the interval [0, 2π]. If we integrate both sides of Equation (1) and assume that it is permissible

1
to integrate the series term-by-term, we get
Z 2π Z 2π ∞ Z 2π ∞ Z 2π
a0
cos (kx) dx + sin (kx) dx,
X X
f (x) dx = dx + ak bk
0 0 2 0 0
k=1 k=1

but cos (kx) dx = sin (kx) dx = 0 because k is an integer. So

R 2π R 2π
0 0
Z 2π
1
a0 = f (x) dx
π 0

To determine an for n = 1, 2, . . . we multiply both sides of Equation (1) by cos(nx) and

integrate term-by-term from 0 to 2π:
Z 2π
f (x) cos(nx) dx
0
∞ ∞
a0 2π
Z  X
 Z 2π Z 2π

cos(nx) cos cos(nx) sin
 X
=  dx + ak (kx) dx + b k (kx) cos(nx)
 dx
2 0  0 0 
 
} k=1 k=1 |
 =0
| {z  {z }
  =0
n−1
X Z 2π  Z 2π ∞ Z 2π 


cos(kx) cos(nx) cos cos cos(nx)
 dx
X
2
= ak   dx + an (nx) dx + ak (kx)
  
k=1 |
 0  {z } |0 {z } k=n+1  | 0
 {z }
 =0 =π  =0
= an π.

Hence
Z 2π
1
an = f (x) cos(nx) dx for n = 1, 2, 3, . . .
π 0

and, similarly,
Z 2π
1
bn = f (x) sin(nx) dx for n = 1, 2, 3, . . .
π 0

give expressions for the Fourier coefficients.

Remark. The integration interval [0, 2π] can be changed by [−π, π]. Consequently, a 2π-periodic
even function verifies all bn coefficients are null, then its Fourier series is a series of cosines. In
the other hand, a 2π-periodic off function verifies all an are null, then its Fourier series is a
series of sines.
Notice that the above arguments used to derive these Fourier coefficients are not rigorous,
because term to term integration needs the uniform convergence of the series. In fact, the equal-
ity (1) is not true where the function f is not continuous, because the right part is continuous
everywhere assuming uniform convergence.
Moreover, a continuous function g defined in a closed integration interval [0, 2π], its Fourier
coefficients exist always1 . But an extension to 2π-periodic function defined recursively as

if x ∈ [0, 2π)

g(x)

f (x) = g(x − 2π) if x ≥ 2π
g(x + 2π) if x < 0



is not necessary continuous, unless g(0) = g(2π). Obviously there are other ways to extend the
function g to a period one.
For these reasons we need redefine the Fourier series to the cases where the equality (1) is
not true for all values of x.
1
Really g could be defined in any interval of length 2π.

2
Definition 1.2. A bounded function f (x) is piecewise
continuous on a finite interval [a, b] provided there ex-
ists a partition

a = x0 < · · · < x n = b
| | | |
of the interval [a, b] and functions f1 , f2 , . . . , fn contin- a x1 x2 b
uous on [a, b] such that for x not a partition point

f1 (x) x0 < x < x1 ,

.. ..

f (x) = . . (2)


f (x) x
n n−1 < x < xn .

The values of f at partition points x0 , x1 , . . . , xn are undecided by this equation (2). That
implies that f (x) has one-sided limits at each point of a < x < b

f (x+ ) = lim f (t) and f (x− ) = lim f (t)

t→x+ t→x−

and appropriate one-sided limits at the endpoints. Therefore, f has at worst a jump discon-
tinuity at each partition point.

Example 1.3 (square wave function). We are computing the Fourier coefficients and Fourier
series of the following piecewise continuous and 2π-periodic function defined by

0 if − π ≤ x < 0
(
f (x) =
1 if 0 ≤ x < π

1
and period T = 2π. −2π −π 0 π 2π

Using formulas for the Fourier coefficients,

1 2π
Z Z π Z 2π 
1
a0 = f (x) dx = 1 dx + 0 dx = 1
π 0 π 0 π
and for n ≥ 1,
1 2π 1 π 1 sin nx π
Z Z 
an = f (x) cos(nx) dx = cos(nx) dx = =0
π 0 π 0 π n 0
1 2π 1 π 1 cos nx iπ 1 − (−1)n
Z Z
bn = f (x) sin(nx) dx = sin(nx) dx = − =
π 0 π 0 π n 0 nπ

Therefore the Fourier series is

1 2 2 2
S(x) = + sin x + sin 3x + sin 5x + . . .
2 π 3π 5π
∞ ∞
1 X 1−(−1)k 1 X
S(x) = + kπ sin kx = + (2k−1)π sin(2k − 1)x.
2
2 2
k=1 k=1

3
 1 1

−π 0 π 2π −π 0 π 2π

(a) S1 (x) = 1
2
+ 2
π
sin x. (b) S3 (x) = 1
2
+ 2
π
sin x + 2
3π
sin 3x.

1 1

−π 0 π 2π −π 0 π 2π

(c) S5 (x) = 1
2
+ π2 sin x + 3π
2
sin 3x + 5x
2
sin 5x. (d) S11 (x) = 1
2
+ 2
π
sin x + · · · + 2
11π
sin 11x.

Figure 2: Here some graphics for Example 1.3.

Remark. Observe in graphs of figures 2 that near discontinuous points of function f the values of
the function and the values of the series are not equals. The nth partial sum of the Fourier series
has large oscillations near the jump. This behavior was observed by experimental physicists,
and it receives the name of Gibbs phenomenon.

Theorem 1.4 (Dirichlet). If f is a periodic function with period 2π and f and f 0 are piecewise
continuous on [0, 2π], then the Fourier series is convergent.
The sum of the Fourier series is equal to f (x) at all numbers where f is continuous.
At the numbers x where f is not continuous, the sum of the Fourier series is the average of
both lateral limits, that is

f (x+ ) + f (x− )
2
We use the notation
∞
a0 X
f (x) ∼ + ak cos (kx) + bk sin (kx)
2
k=1

for representing this situation. Symbol ∼ means = for x such that f (x) is continuous, but it is
not true for discontinuity points.

2 Fourier Series for Other Periodic Functions

We can find its Fourier series by making a change of variable. In engineering It is usual use the
real variable t (time) for functions. Suppose f (t) has period T , that is f (t + T ) = f (t) for all t,
and we let x = 2πtT and
 
˜ Tx
f (x) = f
2π

is a function with period 2π and t = ±T corresponds x = ±2π. Indeed,

     
˜ T (x + 2π) Tx Tx
f (x + 2π) = f =f +T =f = f˜(x).
2π 2π 2π

4
So the Fourier series of f (t) can be obtained from the Fourier series of f˜(x):
∞
a0 X
f˜(x) ∼ + ak cos (kx) + bk sin (kx)
2
k=1
∞
a0
ak cos + bk sin
X
2kπt 2kπt



f (t) ∼ + T T
2
k=1

And the Fourier coefficients

1 2π ˜ 1 2π ˜ 2π
Z Z Z
1
a0 = f (x) dx, an = f (x) cos(nx) dx, bn = f˜(x) sin(nx) dx,
π 0 π 0 π 0

changing variable t = Tx
2π
Z T Z T Z T
2 2 2
f (t) cos n 2π f (t) sin n 2π



a0 = f (t) dt, an = T t dt, bn = T t dt.
T 0 T 0 T 0

It is easy to see that it’s possible choose any interval [a, a + T ] instead of [0, T ].
Usually is defined the fundamental frequency ω = 2π T,
∞
a0 X
f (t) ∼ + ak cos (kωt) + bk sin (kωt)
2
k=1

with Fourier coefficients

Z T Z T
2 2
an = f (t) cos(nωt) dt; bn = f (t) sin(nωt) dt
T 0 T 0

2.1 Term-by-term differentiation and integration.

Theorem 2.1. Let f (t) be a continuous function on [0, T ] such that f (0) = f (T ) and suppose
also that f 0 (t) is piecewise continuous. Then for all t strictly inside the interval, i.e. t ∈ (0, T ),
at which f 0 (t) exists, the derivative can be obtained term-by-term differentiation of the Fourier
series representation of f (t). So, if has the Fourier series representation
∞
a0 X 2π
f (t) = + ak cos (kωt) + bk sin (kωt) for ω = and 0 ≤ t ≤ T
2 T
k=1

then, if f 0 (t) exists, its Fourier is

∞
−ak k sin (kωt) + bk k cos (kωt) for 0 < t < T.
X
0
f (t) = ω
k=1

Remark. Exercise ?? proves that condition f (0) = f (2π) is necessary.

Theorem 2.2 (Termwise integration). The integral of any function f (t) satisfying Theorem 1.4
on the interval [0, T ] can be obtained by term-by-term integration of its Fourier series represen-
tation. So, if
∞
a0 X
f (t) = + ak cos (kωt) + bk sin (kωt) for 0 ≤ t ≤ T
2
k=1

then
Z t
a0 X ak sin (kωt) − bk (cos (kωt) − 1)
∞
F (t) = f (u) du = t + for 0 ≤ t ≤ T.
o 2 kω
k=1

5
 1

−4 −3 −2 −1 0 1 2 3 4

−1

Figure 3: Example 2.3.

Remark. The above series representation of F (t) is not a Fourier series.

Example 2.3 (Triangle wave function). Find the Fourier series of the function defined by
f (t) = |t| if − 1 ≤ t ≤ 1.
Extending f (t) to periodic of period 2 and ω = π. Choosing interval [−1, 1] we calculate the
Fourier coefficients
2 1
Z Z 0 Z 1
a0 = |t| dt = −t dt + dt = 1,
2 −1 −1 0

2 cos (nπ) − 2 if n is even

(
2 1
Z
0
an = |t| cos(nπt) dt = =
2 −1 n2 π 2 −4
n2 π 2
if n is odd,
Z 1
2
bn = |t| sin(nπt) dt = 0.
2 −1
Therefore
1 4 4 4
f (t) = − 2 cos(πt) − 2 cos(3πt) − cos(5πt) − . . . (3)
2 π 9π 25π 2
Note the very fast convergence of the Fourier series. In the graphic of Figure 3 the first three
terms give a very good approximation to the function.
Example 2.4. Using the previous example, for t = 0, we can calculate the same numerical
that Example 4.3
1 1 1 1 π2
1+ + + + + · · · = .
32 52 72 92 8
By differentiation of (3) at t = 1/2, it is easy to see the new series result:
X (−1)k+1 1 1 1 π
=1− + − + ··· =
2k − 1 3 5 7 4
k≥1

By integration of (3), we obtain

Z 1/2 Z 1/2 Z 1/2
1 4 X 1
f (t) dt = dt − 2 (2k−1)2
cos(2k − 1)πt dt
0 0 2 π 0
k≥1
1 1 4 X 1 sin (2k−1) π2


= − 2 (2k−1)2 (2k−1)π
8 4 π
k≥1
1 1 4 X (−1)k+1
= − 3 (2k−1)3
8 4 π
k≥1

Hence
X (−1)k+1 π3
3
= .
(2k − 1) 32
k≥1

6
3 Fourier Sine and Cosine Series on 0 ≤ x ≤ T
Sometimes is useful a representation of a function defined on an interval [0, T ] like a Fourier
series consisting only by sines (or cosines). But these representations are efficient only in that
interval.
Let a function f (t) specified on the interval 0 ≤ t ≤ T .

f (t) f (t)

Sn (t) Sn (t)

−T T −T T

(a) Sine series. (b) Cosine series.

Sine series. We can define a new odd function g(t) given by

−f (−t) if − T ≤ t < 0,
(
g(t) =
f (t) if 0 ≤ t ≤ T

and we make it periodic with period 2T. Then the Fourier series of g(t) consists only by sines,
and it represents f (t) uniquely on the interval (0, T ].
Remark. In case of f (0) 6= 0, the function g(t) has a jump discontinuity in 0 and it produces
the Gibbs phenomenon.

Cosine series. We can define a new even function g(t) given by

f (−t) if − T ≤ t < 0,
(
g(t) =
f (t) if 0 ≤ t ≤ T

and we make it periodic with period 2T. Then the Fourier series of g(t) consists only by
cosines, and it represents f (t) uniquely on the interval (0, T ].
Example 3.1. Represent the function f (t) = 3 + 4t − 1 (t − 2) (3t − 3) by sine and cosine

series on interval 0 ≤ t ≤ 4.
Extending f (t) to an odd function on interval [−4, 4], we get the fundamental frequency
ω = 2π8 = 4 , and
π

∞
336−6π 2 n2 − 6π 2 n2 −240 (−1)n


Sine series: f (t) = sin n π4 t on 0 ≤ t ≤ 4.
X

3
π n 3
n=1
∞
576−84π 2 n2 + 36π 2 n2 −576 (−1)n


Cosine series: f (t) = 2 + cos n π4 t on 0 ≤ t ≤ 4.
X

4
π n4
n=1

4 Complex Notation
By using the complex notation for sine and cosine functions,
eiφ + e−iφ eiφ − e−iφ
cos φ = , sin φ =
2 2i

7
we may write the formula for the Fourier series in a more compact way:
∞
a0 X eikωt + e−ikωt eikωt − e−ikωt
f (t) ∼ + ak + bk =
2 2 2i
k=1
∞   ∞  
a0 X ak bk X ak bk
= + −i e ikωt
+ +i e−ikωt
2 2 2 2 2
k=1 k=1

calling c0 = a20 , ck = a2k − i b2k for k > 0 and ck = for k < 0, function f (t) could be
a−k b−k
2 +i 2
written in a more compact way
∞ Z T
1
with cn =
X
f (t) ∼ ck e ikωt
f (t)e−inωt dt
T 0
k=−∞

This is called the complex Fourier series. Please note that the summation now also covers
negative indexes, we have “negative frequencies”.
Remark. From now, we will consider integrals over complex functions defined on real intervals,
i.e. if ξ(t) is complex function on a ≤ t ≤ b, we will understand the integral as the integrals
over each one of its real and imaginary part.
Z b Z b Z b
ξ(t) dt = Re(ξ(t)) dt + i Im(ξ(t)) dt.
a a a

Example 4.1. For the complex notation for the Fourier series expansion of the triangle wave
function (defined in Example 2.3)

f (t) = |t| if − 1 ≤ t ≤ 1 and f (t + 2) = f (t) for all t.

we do
1
(−1)n − 1 1
Z Z
1 1 1
cn = |t|e−inπt dt = 2 2
, for n ∈ Z, n 6= 0 and c0 = |t| dt = .
2 −1 π n 2 −1 2
Therefore
−1 ∞
1 X ((−1)n − 1) eiπ n t 1 1 X ((−1)n − 1) ei π n t
f (t) = 2 + + 2 =
π n=−∞ n2 2 π n2
n=1
= ··· − 2
25π 2
e−i5πt − 2 −i3πt
9π 2
e − 2 −iπt
π2
e + 12 − 2 iπt
π2
e − 2 i3πt
9π 2
e − 2
25π 2
ei5πt − ···

Separating the real and imaginary parts, the previous expression (3) is obtained.

4.1 Parseval’s Identity

Formal proof of the Parseval’s identity is laborious and requires prior knowledges not covered in
this course. In spite of that, if we admit some hypotheses about the convergence of the Fourier
series we can offer an informal justification.
Assuming f continuous and T -periodic,
∞
X
f (t) = ck eikωt
k=−∞

Hence
Z T Z T ∞
X ∞
X ∞
X Z T
ikωt −imωt
f (t)f (t) dt = ck e cm e dt = ck cm ei(k−m)ωt dt,
0 0 k=−∞ m=−∞ k,m=−∞ 0

8
but
Z T Z T Z T
e i(k−m)ωt
dt = cos(k − m)ωt dt + i sin(k − m)ωt dt = 0 for k 6= m
0 0 0

and then
Z T ∞
X ∞
X
f (t)f (t) dt = ck ck T = T |ck |2
0 k=−∞ k=−∞

Theorem 4.2 (Parseval’s Identity). Let f (t) be a continuous function at interval [0, T ] and
periodic with period T , then
Z T ∞
1 2
X
|f (t)| dt = |ck |2 ,
T 0 k=−∞

being cn its complex Fourier coefficients, or

T ∞
a2 X 2
Z
2 2
|f (t)| dt = 0 + ak + b2k ,
T 0 2
k=1

being a0 , an , bn its real Fourier coefficients.

Example 4.3. Applying the Parseval’s identity to Fourier series obtained in Example 1.3 above,
we found value of a numerical series
∞  
2 1 X 2  1−(−1)k 2
π= + 0 + kπ
2π 2
k=1
or
∞
1 4 X 1
= 2
2 π (2k − 1)2
k=1
∞
X 1 π2
2
=
(2k − 1) 8
k=1

Example 4.4. For function in Example 4.1, the Parseval’s identity says:
Z 1 −1 ∞
1 X ((−1)n −1)2 1 X ((−1)n −1)2
|t|2 dt = π 4 n4
+ + π 4 n4
2 −1 n=−∞
4
n=1
∞
1 2 1 2 X ((−1)n −1)2
· = + 4 n4
2 3 4 π
n=1

and finally
∞
X ((−1)n − 1)2 4 4 4 π4
= + + + · · · =
n4 14 34 54 24
n=1

or what is the same thing

∞
X 1 1 1 1 π4
= + + + · · · = .
(2n − 1)4 14 34 54 96
n=1