FOURIER SERIES

Fourier series are important for solving pdes. We discuss the

basic properties of this kind of series here. The application
will be discussed later.

Expanding a function in terms of sine and cosine

Supposed that f (x) is a function defined on the interval [−π, π].

Let us try to represent f (x) by the following expression.
∞
1 X
f (x) = a0 + [an cos(nx) + bn sin(nx)] (1)
2 n=1

This may or may not work. If it works, what would the coef-
ficients be? We can proceed heuristically as following. First,
introduce the notations

Xn(x) = cos(nx), Yn(x) = sin(nx), n = 0, 1, 2, ... (2)

Z π
(f, g) = f (x) g(x) dx. (inner product) (3)
−π

The expression (1) can also be written as

∞
X
f (x) = [cn Xn(x) + bnYn(x)] (4)
n=0

a0
with c0 = and cn = an for n = 1, 2, ....
2
1
Recall the formulas
1
sin(mx) cos(nx) = [sin(m − n)x + sin(m + n)x],
2
1
cos(mx) cos(nx) = [cos(m − n)x + cos(m + n)x],
2
1
sin(mx) sin(nx) = [cos(m − n)x − cos(m + n)x],
2
so that
(X0 , X0) = 2π,
and for m, n = 1, 2, ...
(Xm , Yn) = 0,
(Xm , Xn) = πδmn,
(Ym , Yn) = πδmn,

0 for m 6= n
where δmn =
1 for m = n
is the Kronecker delta.
The functions Xn, Ym are saidpto be orthogonal to each other.
The norm of Xn is kXnk = (Xn, Xn).

Multiply both sides of (4) by X0(x) and integrate both sides

over [−π, π] to get
Z π
a0 (X0 , f ) 1
c0 = = = f (x)dx. (5)
2 (X0 , X0) 2π −π

Repeating the procedure with Xm for m = 1, 2, 3, ..., one gets

2
 π
(Xm, f ) 1
Z
cm = am = = f (x) cos(mx). (6)
(Xm , Xm) π −π

Similarly,
π
(Ym , f ) 1
Z
bm = = f (x) sin(mx). (7)
(Ym , Ym) π −π

The steps of derivation are flawed by the interchange of inte-

gration and infinite summation. Nevertheless, this calculation
shows how the coefficients should be chosen when the inter-
change is permissible. We will show that for functions having
certain properties, the result turns out to be correct.
Definition:
The Fourier series of a function f (x) on [−π, π] is defined
to be
∞
1 X
a0 + [an cos(nx) + bn sin(nx)]
2 n=1
in which the coefficients are given by equations (5), (6),
and (7).

Example:
Find the Fourier
 series for
−1/2 for −π ≤ x < 0
f (x) =
1/2 for 0 ≤ x ≤ π.
∞
X 1
(2/π) sin(2n − 1)x
n=1
2n − 1

N ote convergence problems at jumps

3
Change of scale

Definition:
A function f is said to be periodic with period P (6= 0) if
f (x + P ) = f (x)
for all x. The least P is called the fundamental period.

Suppose that f is periodic of period 2L and integrable on

[−L, L]. The periodicity property makes it adequate to con-
sider f only on the interval [−L, L]. The Fourier series of f
on [−L, L] is
∞  nπx   nπx 
1 X
a0 + an cos + bn sin .
2 n=1
L L

The Fourier coefficients

Z L are
1  nπx 
an = f (x) cos dx for n = 0, 1, 2, ...,
L −L L
Z L  nπx 
1
bn = f (x) sin dx for n = 1, 2, ... .
L −L L

Note that any function defined on [−L, L] can be extended

to a periodic function of period 2L over (−∞, ∞).

Half-range Fourier expansions

Suppose that a function g is defined on [−L, L]. We say that

g is an even function if g(−x) = g(x). We say that g is an
odd function if g(−x) = −g(x).
4
Suppose that f is integrable on [0, L]. It can be extended to
create an odd function g defined on [−L, L] by setting

f (x) for 0 ≤ x ≤ L
g(x) =
−f (−x) for −L < x < 0.
For x ∈ [0, L], the Fourier series for g converges to f the same
way as for g. The coefficients an (n = 0, 1, 2...) of the Fourier
series for g all vanishes. The bn coefficients can be computed
as
Z L  nπx 
2
bn = f (x) sin dx.
L 0 L
The expression
X∞  nπx 
bn sin
n=1
L
is called the Fourier sine series of f on [0, L].
On the other hand f can be extended to [−L, L] as an even
function: 
f (x) for 0 ≤ x ≤ L
g(x) =
f (−x) for −L < x < 0
and the Fourier series for g can also be used to describe f on
[0, L]. With
2 L  nπx 
Z
an = f (x) cos dx, n = 0, 1, 2, ...,
L 0 L
the Fourier cosine series of f on [0, L] is
∞  nπx 
1 X
a0 + an cos .
2 n=1
L
5
Note that the sine series gives a 0 value at both x = 0 and
x = L while the cosine series converges to f (0+) and f (L−)
at the two boundary points.

Example:
Find the Fourier cosine series of f (x) = sin(x) on [0, π].
∞
2 4 X 1
− cos(2mx)
π π m=1 4m2 − 1

6
Three notions of convergence

How good is a series expansion for a function? What is the

meaning of
X∞
fn = f ?
n=1

(e.g. fn(x) = bn sin(nx) in a Fourier sine series)

1. Pointwise convergence
Definition:
P∞
An infinite series n=1 fn(x) is said to converge to
f (x) pointwise in (a, b) if it converges to f (x) for each
x ∈ (a, b). That is
N
X
f (x) − fn(x) → 0 as N → ∞.
n=1

2. Uniform convergence
The uniform norm of a real- or complex-valued function
f defined on a set S is the nonnegative number
kf k∞ = sup{|f (x)| : x ∈ S}.
It is also called the supremum norm.

7
 Definition:
The series is said to converge uniformly to f (x) if
N
X
f− fn → 0 as N → ∞.
n=1 ∞

3. L2 convergence
Definition:
The series is said to converge in the mean-square (or
L2 ) sense to f (x) if
Z b XN 2
f (x) − fn(x) dx → 0 as N → ∞.
a n=1

The integral in this definition is in the Lebesgue sense. For

this course, most of the time, we can think of it as a
Riemann integral for piecewise continuous functions.
Definition:
A function f (x) is said to have a jump at x0 if both
one-sided limits lim+ and lim exist [denoted as f (x+ 0)
x→x0 x→x−
0
and f (x−
0 ) respectively] but not equal.

Definition:
A function is said to be piecewise continuous if in each
finite interval it has only a finite number of jumps and it
is continuous at all other points.
8
Uniform convergence is the strongest

(i) Pointwise convergence does not imply uniform

convergence.

Example:
n n−1
Let fn(x) = − on (0, 1).
1 + n2 x2 1 + (n − 1)2 x2
N
X N
The N-th partial sum SN (x) = fn = 2 x2
n=1
1 + N
converges to f (x) = 0 in the pointwise sense.
N
However, 2 2
− f (x) = N grows with N .
1+N x ∞

(ii) Pointwise convergence does not imply mean-square

convergence.

Example:
The mean-square distance between SN and f (x) of
previous example is
Z 1 2 Z N
N 1 2
2 x2
− f (x) dx = N ( 2
) dy.
0 1 + N 0 1 + y
It does not → 0.

9
(iii) Mean-square convergence does not imply pointwise
convergence.

Example:
On (0, 1), let
0 for x 6= 12

fn(x) = 1 1
2n for x = 2 .
The N-th partial sum is
N
for x 6= 12

X 0
SN (x) = fn(x) =
1 − ( 12 )N for x = 12 .
n=1
Z 1 2
As SN (x) − 0 dx = 0, the series converges to
0
f (x) = 0 in the mean-square sense, but not in the
pointwise sense.

(iv) Uniform convergence ⇒ pointwise convergence

N
X XN
f (x) − fn(x) ≤ f (x) − fn(x)
n=1 n=1 ∞

(v) Uniform convergence ⇒ mean-square convergence

Z b N
X 2 N
X 2 Z b
f (x) − fn(x) dx ≤ f (x) − fn(x) dx
a n=1 n=1 ∞ a

10
Furthermore, uniform convergence allows for integration and
differentiation across summation. The following are useful
theorems from an analysis course.
Theorem:
(Dominated convergence) Let fn, f , g be piecewise
continuous functions on [a, ∞) such that (i) for every
b > a, fn → f uniformlyRon [a, b], (ii) for all xR > a,
∞ ∞
|fn(x)|
R∞ ≤ g(x),
R∞and (iii) a
g is finite. Then a
f exists,
and a fn → a f as n → ∞.
Theorem:
(Differentiation across summation)
Suppose (i) each fn is differentiable and fn0 is continuous
on (a, b), (ii) fn(c) is convergent for some c ∈ (a, b), and
(iii) fn0 is uniformly convergent to φ on (a, b). Then on
(a, b), fn is uniformly convergent to a differentiable
function f and f 0 = φ.
The corresponding theorem for improper integrals will be
used several times later.
Theorem:
(Differentiation across an improper integral)
Let f (x, t) and ∂f /∂t be continuous functions
R ∞ on
(a, ∞)
R ∞× (c, d). Assume that the integrals a |f (x, t)|dx
and a |∂f /∂t(x, t)|dx converges uniformly (as
improper integrals) for t ∈ (c, d). Then
Z ∞ Z ∞
d ∂f
f (x, t)dx = (x, t)dx for t ∈ (c, d).
dt a a ∂t
11
Mean square distance between f and its Fourier
series

The mean square distance between f and the N -th partial

sum of the Fourier series expansion is
XN XN
EN2 = (f − [cn Xn + bnYn], f − [cnXn + bnYn])
n=0 n=0
XN XN
= (f, f ) − 2 (f, cn Xn + bnYn) + [c2nkXnk2 + b2nkYnk2 ]
n=0 n=0
XN
= kf k2 − [c2nkXnk2 + b2nkYnk2] ≥ 0,
n=0

as the mean square distance cannot be less than 0. The sum

of the squares of the coefficients are bounded by the
inequality
XN
[c2n kXnk2 + b2nkYnk2] ≤ kf k2
n=0
This is called Bessel’s inequality. Similar inequalities hold
for Fourier sine and Fourier cosine expansions.
The Fourier series converges to f in the mean-square sense if
and only if equality holds. In such a case the expression is
call Parseval’s equality.

The following theorem, stated here without proof, points out

that any square-integrable function is equal to its Fourier
series almost everywhere.
12
Theorem:
(L2 Convergence) The Fourier series converges to f in the
mean-square sense in (a, b) provided only that f (x) is
any function for which
Z b
2
||f || = |f (x)|2 dx is finite (in the Lebesgue sense).
a

As a consequence, the Parseval’s equality holds.

13
Appendix

Proof of theorem on integration across summation

(Dominated Convergence)

Proof:

RLet  > 0. By condition (iii) one can choose b > a with

∞
b
g < , and by condition (i) there is an N such that
|fn(x) − f (x)) < /(b − a) for n > N and x ∈ [a, b].
Then
Z ∞ Z ∞ Z b Z ∞
fn − f ≤ |fn − f | + |fn − f |
a a a b
Z b Z ∞
≤ /(b − a) + 2 g = 3.
a b

Proof of theorem on differentiation across summation

 
d d
( lim fn(x)) = lim fn(x)
dx n→∞ n→∞ dx
Proof:
Since fn0 is uniformly convergent to φ, φ is continuous fn
can be written as Z x
fn(x) = fn(c) + fn0 (ξ)dξ
c
Z x Z x
= fn(c) + [fn0 (ξ) − φ(ξ)]dξ + φ(ξ)dξ.
c c
14
 For any  > 0, choose N so that |fn0 (ξ) − φ(ξ)| <  for all
ξ ∈ (a, b) and n ≥ N . Then
Z x
fn(x) − fn(c) − φ(ξ)dξ < (b − a).
c
Thus fn is uniformlyZconvergent to
x
f (x) = lim fn(c) − φ(ξ)dξ, and f 0 = φ.
n→∞ c

Note: Proof of the improper integral version is similar.

15