Fourier Series
Roughly speaking, a Fourier series expansion for a function is a representation of
the function as sum of sins and cosines. Expressing a musical tone as a sum of a fundamental
tone and various harmonics is such a representation. So is a spectral decomposition of light
waves. The main Fourier series expansions that we use in this course are stated in the next
section. We shall never prove that these expansions are correct. But in the section Validity
of Fourier series we give an elementary partial argument that, hopefully, will convince you
that the expansions are indeed correct. In the section Usefulness of Fourier Series we
introduce one of the many ways that Fourier series are used in applications.
The Main Fourier Series Expansions.
We shall shortly state three Fourier series expansions. They are applicable to functions that are piecewise continuous with piecewise continuous first derivative. In applications,
most functions satisfy these regularity requirements. We start with the definition of piecewise continuous.
A function f (x) is said to be piecewise continuous if it is continuous except for
isolated jump discontinuities. In the example below, f (x) is continuous except for jump
f (x)
1
[f (1+)
2
f (1+)
+ f (1)]
f (1)
f (1)
x
1
2
3
discontinuities at x = 1 and x = 2.5. If a function f (x) has a jump discontinuity at x0 , then
the value of f (x) as it enters and leaves the jump are
f (x0 ) = xx
lim f (x)
0
x<x0
and
f (x0 +) = xx
lim f (x)
0
x>x0
respectively. If f were continuous at x0 , we would have f (x0 ) = f (x0 +) = f (x0 ). At
1
a jump, however, there is no a priori relation between f (x0 ) and f (x0 ). In the example
above, f (1) is well below both f (1) and f (1+). On the other hand, it is fairly common
for the value of f at the jump x0 to be precisely at the midpoint of the jump. That is
f (x0 ) = 21 [f (x0 +) + f (x0 )]. In the example, this is the case at x0 = 2.5.
Theorem (Fourier Series) Let f (x) be piecewise continuous with piecewise continuous first
derivative.
a) Let f (0) = f (`) = 0. Then
bk sin
k=1
kx
`
f (x)
if f is continuous at x
f (x+)+f (x)
2
otherwise
R`
2
`
for all 0 x ` if and only if bk =
b) Let f 0 (0) = f 0 (`) = 0. Then
a0
2
kx
ak cos
k=1
=
2
`
for all 0 x ` if and only if ak =
c) Let f be periodic of period 2`. Then
a0
2
X
ak cos
k=1
kx
`
for all x if and only if
Z `
1
f (x) cos
ak = `
kx
`
dx .
if f is continuous at x
f (x+)+f (x)
2
otherwise
R`
0
f (x)
dx
kx
`
f (x) cos
kx
+ bk sin
kx
`
f (x) sin
dx .
f (x)
if f is continuous at x
f (x+)+f (x)
2
otherwise
and
bk =
1
`
f (x) sin
kx
`
dx
Remark. One consequence of the above theorem is that if you are told that, for example,
k=1
k sin
kx
`
=0
for all 0 x `
then, by part a with f (x) = 0, it is necessary that
Z `
2
k = `
f (x) sin kx
dx = 0
`
for all k = 1, 2,
This is used repeatedly in using Fourier series to solve differential equations.
2
Example. Consider the function
f (x)
1
f (x) =
1 if 0 x 1
=
0 if 1 < x 2
1
Apply part b of the Fourier series Theorem with this f and ` = 2. Then
ak =
2
`
f (x) cos
0
kx
`
dx =
cos
0
kx
2
dx =
1
1
if k = 0
= 0
k
2
sin
if
k
k
2
2
(1)(k1)/2 k
if k = 0
2
k
1
kx
sin 2
0
if k 1
if k = 0
if k > 1 and k is even
if k is odd
and by part b of the Theorem
1+
2
(1)p (2p+1)
cos
p=0
(2p+1)x
2
if 0 x 1
=
if x = 1
if 1 < x 2
1
2
We have summed over all odd k by setting k = 2p + 1 and summing over p = 0, 1, 2, .
PN
2
There is a Java demo that shows the graphs of 1 + p=0 (1)p (2p+1)
cos (2p+1)x
for
2
many values of N .
Usefulness of Fourier Series.
In this course, we use Fourier series as a tool for solving partial differential equations.
This is just one of its many applications. To give you a preliminary taste of this application,
we now briefly consider one typical partial differential equation problem. It arises in studying
a vibrating string. Suppose that a vibrating string has its ends tied to nails at x = y = 0
and x = `, y = 0. Denote by y(x, t) the amplitude of the string at position x and time t. If
3
there are no external forces, no damping and the amplitude is small,
2y
t2 (x, t)
y
= c2 x
2 (x, t)
for all 0 x `, t 0
(1)
y(0, t) = 0
for all t 0
(2)
y(`, t) = 0
for all t 0
(3)
y(x, 0) = f (x)
for all 0 x `
(4)
for all 0 x `
(5)
y
t (x, 0)
= g(x)
A detailed derivation of this system of equations is provided in another set of notes. The first
equation is the Newtons law of motion appropriate for the current situation; the next two
equations impose the requirements that the ends of the string are tied to nails; the final two
equations specify the initial position and and speed of the string. We are assuming that the
initial position and speed f (x) and g(x) are given functions. The unknown in the problem is
the amplitude y(x, t). For each fixed t 0, y(x, t) is a function of the one variable x. This
function vanishes at x = 0 and x = ` and thus, by part a) of the Fourier series theorem has
an expansion
y(x, t) =
kx
`
bk (t) sin
k=1
The solution y(x, t) is completely determined by the, as yet unknown, coefficients bk (t).
P
kx
Furthermore these coefficients can be found by substituting y(x, t) =
k=1 bk (t) sin
`
into the above five requirements on y(x, t). First the PDE (1):
0=
2y
t2 (x, t)
2y
c2 x
2 (x, t)
=
=
b00k (t) sin
k=1
h
X
b00k (t)
k=1
kx
`
k 2 2 c2
`2 bk (t) sin
k=1
k 2 2 c2
`2 bk (t)
sin
kx
`
kx
`
This says that, for each fixed t 0, the function 0, viewed as a function of x, has Fourier
i
P h
2 2 2
series expansion k=1 b00k (t) + k `2 c bk (t) sin kx
. Our Fourier series theorem then forces
`
b00k (t) +
k 2 2 c2
`2 bk (t)
=0
for all k, t
(10 )
Because sin 0 = sin k = 0 for all integers k, conditions (2) and (3) are satisfied by y(x, t) =
4
k=1 bk (t) sin
kx
`
regardless of what bk (t) are. Finally substituting into (4) and (5) gives
y(0, t) =
y
(0, t)
t
k=1
bk (0) sin
kx
`
= f (x)
b0k (0) sin
kx
`
= g(x)
k=1
By uniqueness of Fourier coefficients, once again,
Z `
2
bk (0) = `
f (x) sin
0
b0k (0)
2
`
kx
`
g(x) sin
0
kx
`
dx
(40 )
dx
(50 )
For each fixed k, equations (1), (4) and (5) constitute one second order constant coefficient
ordinary differential equation and two initial conditions for the unknown function bk (t). You
already know how to solve constant coefficient ordinary differential equations. Denoting
Z `
Z `
kx
2
2
f (x) sin ` dx
g(x) sin kx
k = `
k = `
dx
`
0
the solution of (1), (4) and (5) is
bk (t) = k cos
kct
`
and the solution of (1), (2), (3), (4), (5) is
y(x, t) =
h
X
k cos
k=1
kct
`
`k
kc
`k
kc
sin
sin
kct
`
kct
`
i
sin
kx
`
The interpretation of this formula is provided in another set of notes.
Validity of Fourier Series.
It is beyond the scope of this course to give a full justification for the Fourier series
theorem. However, with just a little high school trigonometry, we can justify a formula that is
useful in its own right and that also provides arbitrarily good approximations to the Fourier
series expansions. For concreteness, we shall just consider Fourier sin series that is part
a) of the theorem. The other parts are similar. Imagine some application in which we have
5
to measure some function f (x) obeying f (0) = f (`) = 0. For example, f (x) might be the
amplitude at some fixed time of a vibrating string. Because we can only make finitely many
measurements, we cannot determine f (x) for all values of x. Suppose that we measure f (x)
for x =
2`
`
N, N,
1)`
, (N N
where N is some fixed integer. Define
Fn(N ) = f
n`
N
Because we do not know f (x) for all x we cannot compute bk =
2
`
R`
0
f (x) sin
kx
`
dx . But
we can get a Riemann sum approximation to it using only xs for which f (x) is known. All
we need to do is divide the domain of integration up into N intervals each of length `/N . On
(n+1)`
kx
n`
k n`
the interval n`
we
approximate
f
(x)
sin
by
f
sin
N
N
`
N
` N .
f (x) sin kx
`
x
`
N
2`
N
(N 1)`
N
This gives
(N )
Bk
2
`
N
1
X
n`
`
Nf N
n=0
sin
k n`
` N
2
N
N
1
X
Fn(N ) sin
n=0
nk
N
More generally, let F0 , F1 , , , FN be N + 1 numbers obeying F0 = FN = 0.
Define
Bk =
2
N
N
1
X
Fn sin
mk
N
m=0
We shall show that the Fourier sin series representation for the Fn s
Fn =
N
1
X
Bk sin
k=0
nk
N
(1)
is really true. To do so we just evaluate the right hand side by substituting in the definition
6
of the Bk s.
N
1
X
Bk sin
nk
N
k=0
2
N
N
1 N
1
X
X
Fm sin
mk
N
sin
nk
N
Fm sin
mk
N
sin
nk
N
k=0 m=0
2
N
N
1 N
1
X
X
m=0 k=0
N
1
X
Fm
m=0
2
N
N
1
X
mk
sin
k=0
sin
The validity of (1) is then an immediate consequence of part b) of
)
nk
N
Lemma.
a)
N
1
X
cos(k) =
k=0
N
1
2 (1 cos N ) +
1 sin N sin
2 1cos
if cos = 1
if cos 6= 1
b) Let n and m be integers with 1 n, m N 1. Then
2
N
N
1
X
sin
k=0
mk
N
sin
nk
N
1
0
if n = m
if n =
6 m
Proof: a) If cos = 1 then is an integer multiple of 2 so that cos(k) = 1 for all integers
PN 1
PN 1
k and k=0 cos(k) = k=0 1 = N , as claimed.
If cos 6= 0, we must show
N
1
X
cos(k) = 21 (1 cos N ) +
1 sin N sin
2 1cos
k=0
or equivalently (just moving the 12 (1 cos N ) to the left hand side)
1
2
N
1
X
cos(k) +
1
2
cos(N ) =
1 sin N sin
2 1cos
k=1
or equivalently (just cross multiplying by 2(1 cos ))
(1 cos ) +
N
1
X
2(1 cos ) cos(k) + (1 cos ) cos(N ) = sin N sin
k=1
The left hand side is
(1 cos ) +
N
1
X
2(1 cos ) cos(k) + (1 cos ) cos(N )
k=1
= 1 cos +
N
1
X
2 cos(k)
k=1
= 1 cos +
N
1
X
N
1
X
2 cos cos(k) + cos(N ) cos cos(N )
k=1
2 cos(k)
k=1
1
2
cos (k + 1))
k=1
1
2 cos
+ cos(N )
by the trig identity cos cos =
N
1
X
N
1
X
cos (k 1))
k=1
(N 1))
cos( + ) +
1
2
1
2
cos (N + 1))
cos( ). Writing out the three sums on
three separate rows
(1 cos ) +
N
1
X
2(1 cos ) cos(k) + (1 cos ) cos(N )
k=1
= 1 cos
+ 2 cos + 2 cos(2) + + 2 cos((N 2) ) + 2 cos((N 1) )
cos(2) cos((N 2) ) cos((N 1) ) cos(N )
1 cos cos(2) cos((N 2) )
1
2
cos (N 1)
= sin() sin(N )
1
2
cos (N + 1)
1
2
cos((N 1)) + cos(N )
1
2
cos((N +1))
which is exactly the desired identity.
b) Using sin sin =
2
N
1
X
1
2
sin
k=0
cos( )
mk
N
sin
nk
N
1
2
cos( + )
=
N
1
X
cos
k=0
(nm)k
N
N
1
X
cos
k=0
We wish to apply the identity of part a twice, the first time with =
time with =
<
0<
(n+m)
.
N
(n+m)k
N
(nm)
N
and the second
Observe that with 1 n, m N 1 we have
(N 2)
N
(nm)
N
(N 2)
N
< = cos
2
N
(n+m)
N
2(N 1)
N
< 2 = cos
8
(nm)
N
(nm)
N
=1
6= 1
if and only if n = m
and
(n+m)
=
sin
N
=0
sin N (nm)
N
N
cos N (nm)
= cos N (n+m)
= (1)nm
N
N
since (1)m = (1)m . Applying part a twice
N
1
X
cos
(nm)k
N
cos
(n+m)k
N
k=0
N
1
X
k=0
1
2
21 (1)nm
1
2
if n = m
if n =
6 m
1
(1)nm
2
Subtracting gives
N
1
X
k=0
cos
(nm)k
N
N
1
X
cos
k=0
(n+m)k
N
N
0
if n = m
if n =
6 m
which is the desired result.
We remark that trig identities are often cleaner looking and easier to derive using
exponentials of complex numbers. This is indeed the case for Fourier series formulae, which
are just big trig identities. See the notes on complex numbers and exponentials.
