Fourier Series

Much of this course concerns the problem of representing a function as a sum of its dierent frequency
components. 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 ability to isolate the signal of a single
radio or television station from the dozens that are being simultaneously received depends on being able to
amplify certain frequencies and suppress others.
We start our look at the theory of Fourier series with two questions:
Question #1
Which functions f(t) have a representation as a sum of constants times cos(kt)s and sin(kt)s? Since cos(kt)
and sin(kt) can be written in terms of complex exponentials
(1)
using
cos(kt) =
1
2
_
e
ikt
+ e
ikt

sin(kt) =
1
2i
_
e
ikt
e
ikt

and, conversely, e
ikt
can be written in terms of cos(kt)s and sin(kt)s using
e
ikt
= cos(kt) i sin(kt)
it is equivalent to ask which functions f(t) have a representation
f(t) =

k=
c
k
e
ikt
(1)
for some constants c
k
. Because it will save us considerable writing we shall start with this form of the
question and return to sines and cosines later.
First, observe that, for every integer k, e
ikt
= cos(kt) +i sin(kt) is periodic with period 2. So the right
hand side is necessarily periodic of period 2. Unless f(t) is periodic with period 2, it cannot possibly have
a representation of the form (1). We shall shortly state a result that says that, on the other hand, every
suciently continuous (details later) function of period 2 has a representation (1). In this course, we shall
never fully justify this claim. On the other hand, it is fairly easy justify an analogous claim for Discrete
Fourier Series, which is the version of Fourier series for functions f(t) that are only dened for t = n, with
n running over the integers and a xed spacing. This is done in the notes DiscreteTime Fourier Series
and Fourier Transforms. Before giving the detailed answer to this question, we consider
Question #2
Suppose that we know that some specic function f(t) has a representation of the form (1). What are the
values of the coecients c
k
? With a little trickery, we shall be able to answer this question completely and
easily. We wish to solve (1) for the c
k
s. At rst this task looks somewhat daunting because (1) is really a
system of innitely many equations (one equation for each value of t) in innitely many unknowns (the c
k
s).
The trick will allow us to reduce this system to a single equation in any one unknown. Suppose, for example,
that we wish to solve for c
17
. The index 17 has been chosen at random. Then we use the orthogonality
relation that, when k = 17,
_

e
ikt
e
i17t
dt =
_

e
i(k17)t
dt =
1
i(k17)
e
i(k17)t

=
1
i(k17)
_
e
i(k17)
e
i(k17)

= 0 (2)
(1)
See the notes entitled Complex Numbers and Exponentials
February 4, 2007 Fourier Series 1
(because e
i(k17)
/e
i(k17)
= e
i(k17)2
= 1 so that e
i(k17)
= e
i(k17)
for all integers k) to eliminate
all the c
k
s with k = 17 from (1). To do so, take (1), multiply it by e
i17t
and integrate the result from
to . This gives
_

f(t)e
i17t
dt =

k=
c
k
_

e
ikt
e
i17t
dt
Because of (2), all of the terms on the right hand side with k = 17 are zero. Thus
_

f(t)e
i17t
dt = c
17
_

e
i17t
e
i17t
dt = c
17
_

dt = 2 c
17
As promised, this is a single equation in the single unknown c
17
, which we can trivially solve.
c
17
=
1
2
_

f(t)e
i17t
dt
Of course, we could have done the same thing with the integer 17 replaced by any other integer m. This
would have given
c
m
=
1
2
_

f(t)e
imt
dt (3)
We now know all of the Fourier coecients and are ready to return to the answer to question #1.
The Main Fourier Series Expansion.
We shall shortly construct a number of dierent Fourier series expansions that are used for various
dierent classes of functions. For all of these expansions, we are going to restrict our attention to functions
that are piecewise continuous with piecewise continuous rst derivative. In applications, most functions
satisfy these regularity requirements. We start with the denition of piecewise continuous.
A function f(t) is said to be piecewise continuous if it is continuous except for isolated jump disconti-
nuities. In the example below, f(t) is continuous except for jump discontinuities at t = 1 and t = 2.5. If
t
f(t)
1 2 3
f(1+)
f(1)
1
2
[f(1+) + f(1)]
f(1)
a function f(t) has a jump discontinuity at t
0
, then the value of f(t) as it approaches t
0
from the left and
from the right are still welldened. These values are
f(t
0
) = lim
tt
0
t<t
0
f(t) and f(t
0
+) = lim
tt
0
t>t
0
f(t)
respectively. If f were continuous at t
0
, we would have f(t
0
) = f(t
0
+) = f(t
0
). At a jump, however, there
is no a priori relation between f(t
0
), f(t
0
) and f(t
0
+). 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 t
0
to be precisely at
the midpoint of the jump. That is f(t
0
) =
1
2
[f(t
0
+) + f(t
0
)]. In the example, this is the case at t
0
= 2.5.
February 4, 2007 Fourier Series 2
Theorem 1 (Fourier Series) Let f(t) be piecewise continuous with piecewise continuous rst derivative
and also be periodic with period 2. Then

k=
c
k
e
ikt
=
_
f(t) if f is continuous at t
f(t+)+f(t)
2
otherwise
for all < t < if and only if c
k
=
1
2
_

f(t)e
ikt
dt for all integers k.
Example 2 As a rst example, we consider the function f(t) whose graph appears in the gure below.
y
t
3 2

2 3
1
1
According to our main Fourier series theorem
c
k
=
1
2
_

f(t)e
ikt
dt =
1
2
_

0
e
ikt
dt +
1
2
_
0

(1)e
ikt
dt
For k = 0
c
k
=
1
2
_

0
dt
1
2
_
0

dt =
1
2

1
2
= 0
For k = 0
c
k
=
1
2
_
1
ik
e
ikt

0

1
2
_
1
ik
e
ikt

=
i
2k
_
e
ik
1 1 + e
ik()

Since e
ik
and e
ik
are both 1 for k odd and +1 for k even
c
k
=
_

2
k
i if k is odd
0 if k is even
The Fourier series theorem tells us that the graph of

k odd
2
ki
e
ikt
is
y
t
3 2

2 3
1
1
Example 3 The function
f(t) = sin
_
t +

4
_
is periodic of period 2 and so has a Fourier expansion. We could compute the Fourier coecients c
k
by
evaluating the integral
c
k
=
1
2
_

f(t)e
ikt
dt =
1
2
_

sin
_
t +

4
_
e
ikt
dt (4)
But it is much easier to simply observe that
f(t) =
1
2i
_
e
i(t+

4
)
e
i(t+

4
)
_
=
1
2i
e
i

4
e
it

1
2i
e
i

4
e
it
=
1
2i
1

2
(1 + i)e
it

1
2i
1

2
(1 i)e
it
=
1
2

2
(1 i)e
it
+
1
2

2
(1 + i)e
it
(5)
February 4, 2007 Fourier Series 3
This is of the form

k=
c
k
e
ikt
with
c
k
=
_

_
1
2

2
(1 i) if k = 1
1
2

2
(1 + i) if k = 1
0 if k = 1, 1
The Fourier coecients of a periodic function are unique. We showed this in the course of answering Question
# 2. So the right hand side of (5) is THE Fourier expansion for f(t) = sin
_
t +

4
_
. There is no need to
evaluate the integrals in (4). Not only would that be a lot of wasted eort, but we would probably make a
mechanical error along the way and end up with the wrong answer.
Example 4 Suppose that the function
f(t) =

k=
c
k
e
ikt
(6)
is realvalued for all < t < . What conclusions can be drawn concerning the Fourier coecients c
k
? A
number is real if and only if it is the same as its complex conjugate. So f(t) is real if and only if f(t) = f(t).
Substituting in (6) gives that f(t) is real if and only if

k=
c
k
e
ikt
=

k=
c
k
e
ikt
=

k=
c
k
e
ikt
For the second equality, we replaced k by k in the sum. This is legitimate because, as k runs over all of the
integers from to +, k also runs exactly once over all integers. Again by the uniqueness of Fourier
coecients (the only if part of Theorem 1), the two sums are equal for all t if and only if c
k
= c
k
for all
k. So we conclude that
f(t) is real for all < t < c
k
= c
k
for all < k <
For example
sin t =
1
2i
_
e
it
e
it

k=
c
k
e
ikt
with c
k
=
_
1
2i
if k = 1

1
2i
if k = 1
0 otherwise
is a realvalued function and c
1
and c
1
are indeed complex conjugates of each other.
Example 5 The function
f(t) =

k=
c
k
e
ikt
(7)
is necessarily periodic of period 2. Suppose that it is also periodic of period . For example, the function
cos(2t) is periodic of period 2 and also periodic of period . What conclusions can be drawn concerning the
Fourier coecients c
k
? The function f(t) has period if and only if f(t) = f(t + ) for all t. Substituting
in (7) gives that f(t) has period if and only if

k=
c
k
e
ikt
=

k=
c
k
e
ik(t+)
=

k=
_
c
k
e
ik
_
e
ikt
February 4, 2007 Fourier Series 4
Once again by the uniqueness of Fourier coecients, the two sums are equal for all t if and only if c
k
= c
k
e
ik
for all k. For k even, e
ik
= 1 and c
k
= c
k
e
ik
is trivially true. But for k odd, e
ik
= 1 and c
k
= c
k
e
ik
is
true if and only if c
k
= 0. So we conclude that
f(t) has period c
k
= 0 for all odd k
For example
cos(2t) =
1
2
_
e
i2t
+ e
2it

k=
c
k
e
ikt
with c
k
=
_
1
2
if k = 2
0 otherwise
does indeed have c
k
= 0 for all odd k.
Using Theorem 1, we can easily come up with lots of variations. In these variations, we shall assume
that all functions are piecewise continuous with piecewise continuous rst derivative and we shall also assume
that f(t) =
f(t+)+f(t)
2
for all t.
Variation #1 period 2
It is easy to modify our Fourier series result to apply to functions that have period 2, for some > 0,
rather than 2. Just rename the variable t in the Fourier Series Theorem to and then make the change of
variables =

t. This gives
f(t) =

k=
c
k
e
ik

t
with c
k
=
1
2
_

f(t)e
ik

t
dt (8)
As a check, note that since e
ik

(t+2)
= e
ik

t
e
i2k
= e
ik

t
, so that e
ik

t
has period 2. The formula for c
k
in (8) can be derived directly, using
_

e
ik

t
e
im

t
dt =
_
2 if k = m
0 if k = m
and the same strategy as led to (3).
Variation #2 sins and coss
To convert (8) into sins and coss just sub in e
ik

t
= cos(
kt

) + i sin(
kt

).
f(t) =

k=
c
k
e
ik

t
=

k=
c
k
_
cos(
kt

) + i sin(
kt

The k = 0 term in this sum is

c
0
_
cos(
0t

) + i sin(
0t

= c
0
The k = 1 and k = 1 terms together are
c
1
_
cos(
t

) + i sin(
t

+ c
1
_
cos(
t

) + i sin(
t

= [c
1
+ c
1
] cos(
t

) + [ic
1
ic
1
] sin(
t

)
since cos(
t

) = cos(
t

) and sin(
t

) = sin(
t

). Similarly, the k = 2 and k = 2 terms together are

c
2
_
cos(
2t

) + i sin(
2t

+ c
2
_
cos(
2t

) + i sin(
2t

= [c
2
+ c
2
] cos(
2t

) + [ic
2
ic
2
] sin(
2t

)
February 4, 2007 Fourier Series 5
and so on. Hence
f(t) = c
0
+ [c
1
+ c
1
] cos(
t

) + [ic
1
ic
1
] sin(
t

) + [c
2
+ c
2
] cos(
2t

) + [ic
2
ic
2
] sin(
2t

) +
It is conventional to rename c
0
to
a0
2
. The extra
1
2
will make some later formulae cleaner. It is also
conventional to rename [c
1
+ c
1
] to a
1
, [c
2
+ c
2
] to a
2
, etc. and [ic
1
ic
1
] to b
1
, [ic
2
ic
2
] to b
2
, etc.
Note that nobody said that c
2
or c
2
had to be a real number. So it is perfectly possible for b
2
to be a real
number. In fact, it usually is. With these new names
f(t) =
a0
2
+

k=1
_
a
k
cos
_
kt

_
+ b
k
sin
_
kt

_
(9)
The coecients (including a
0
, because a
0
= 2c
0
) are determined by
a
k
= c
k
+ c
k
=
1
2
_

f(t)
_
e
ik

t
+ e
ik

dt =
1

f(t) cos
_
kt

_
dt
b
k
= i[c
k
c
k
] =
1
2
_

f(t) i
_
e
ik

t
e
ik

dt =
1

f(t) sin
_
kt

_
dt
(10)
Variation #3 f odd
If f(t) is an odd function, that is if f(t) = f(t) like sin t, then f(t) cos
_
kt

_
is an odd function and
a
k
=
1

f(t) cos(
kt

_
dt = 0
for all k. Also f(t) sin
_
kt

_
is an even function so that
b
k
=
1

f(t) sin(
kt

_
dt =
2

_

0
f(t) sin(
kt

_
dt
So if f has period 2 and is also odd, our Fourier series expansion simplies to
f(t) =

k=1
b
k
sin
_
kt

_
with b
k
=
2

_

0
f(t) sin
_
kt

_
dt (11)
Variation #4 f even
If f(t) is an even function, that is if f(t) = f(t) like cos t, then f(t) sin
_
kt

_
is an odd function and
b
k
=
1

f(t) sin(
kt

_
dt = 0
for all k. Also f(t) cos
_
kt

_
is an even function so that
a
k
=
1

f(t) cos(
kt

_
dt =
2

_

0
f(t) cos(
kt

_
dt
So if f has period 2 and is also even, our Fourier series expansion simplies to
f(t) =
a0
2
+

k=1
a
k
cos
_
kt

_
with a
k
=
2

_

0
f(t) cos(
kt

_
dt (12)
February 4, 2007 Fourier Series 6
Example 6 Consider once again the function of Example 2. The graph of that function is repeated in the
gure below. This function has period 2, takes the value 1 for < t < 0 and the value +1 for 0 < t < 1.
y
t
3 2

2 3
1
1
This function is also odd, so it has a Fourier sin series expansion (11) with = . The Fourier coecient
b
k
=
2

_

0
f(t) sin(
kt

_
dt =
2

_

0
f(t) sin(kt) dt =
2

_

0
sin(kt) dt =
2
k
cos(kt)

0
=
2
k
_
(1)
k
1

=
_
4
k
if k is odd
0 if k is even
The Fourier series expansion for the function f(t) graphed above is
f(t) =

k=1
b
k
sin(kt) =

k=1
k odd
4
k
sin(kt)
One rst observation to make is that whenever t is at a jump discontinuity of f(t), i.e. whenever t is an
integer multiple of , then sin(kt) = 0 for all k and so

k=1
k odd
4
k
sin(kt) = 0, right in the middle of the jump,
as it is supposed to be. To give some idea of how good the Fourier series expansion works, I have graphed
below a number of partial sums S
N
(t) =

1kN
k odd
4
k
sin(kt). The rst graph is of S
1
(t) =
4

sin(t) and is
not a very good likeness of f(t). The second, S
3
(t) =
4

sin(t) +
4
3
sin(3t), is already starting to look a little
like f(t). As we add more and more terms the graphs start looking more and more like f(t), except that
y
t
3 2

2 3
1
1
S
1
(t)
y
t
3 2

2 3
1
1
S
3
(t)
y
t
3 2

2 3
1
1
S
5
(t)
February 4, 2007 Fourier Series 7
y
t
3 2

2 3
1
1
S
19
(t)
y
t
3 2

2 3
1
1
S
59
(t)
they exhibit a little ringing right at the discontinuities. This ringing is always present in partial sums
of Fourier series at jump discontinuities. It is called the Gibbs phenomenon.
Example 7 As another example, we replace the jump discontinuity of the rst example by a ramp. We
shall see that, even if the ramp is moderately steep, the ringing of Gibbs phenomenon disappears. Just to
put in an additional change, Ill make the function even rather than odd. Its graph is given in the gure
below. This function has period 2, takes the value 1 for 0 < t < a and decreases from 1 down to 0 and t
runs from a up to . For 0 t , the function is given by the formula
f(t) =
_
1 if 0 t a
t
a
if a t
This function is also even, so it has a Fourier cosine series expansion (12) with = . The Fourier coecient,
y
t
3 2
a
2 3
1
for k = 0, is
a
k
=
2

_

0
f(t) cos(
kt

_
dt =
2

_

0
f(t) cos(kt) dt =
2

_
a
0
cos(kt) dt +
2

_

a
t
a
cos(kt) dt
To evaluate these integrals we need the indenite integrals of cos(kt) and t cos(kt). The rst one is trivial
_
cos(kt) dt =
1
k
sin(kt) + C
The second is normally computed using integration by parts. But it is easier to just apply
d
dk
to both sides
of _
sin(kt) dt =
1
k
cos(kt) + C
Note that, while we only need this integral for integer k, it is valid for all nonzero k. So it is legitimate to
dierentiate with respect to k. Also note that, while the constant C is independent of t, it is allowed to
February 4, 2007 Fourier Series 8
depend on k, so its derivative with respect to k need not be zero. In any event, applying
d
dk
gives
_
t cos(kt) dt =
t
k
sin(kt) +
1
k
2
cos(kt) + C

so that
a
k
=
2
k
sin(kt)

a
0
+
2
k(a)
sin(kt)

a

2
(a)
_
t
k
sin(kt) +
1
k
2
cos(kt)

a
=
_
2
k

2
k(a)

sin(ka)
2(1)
k
k
2
(a)
+
2
(a)
_
a
k
sin(ka) +
1
k
2
cos(ka)

For k = 0,
a
0
=
2

_

0
f(t) cos(
0t

_
dt =
2

_

0
f(t) dt
The integral is the area under the graph for 0 t . The region under the graph consists of a rectangle
of height 1 and base a and a triangle of height 1 and base a. So
a
0
=
2

_
a +
1
2
( a)

= 1 +
a

The Fourier series expansion for the function f(t) graphed above is
f(t) =
a0
2
+

k=1
a
k
cos(kt)
with the a
k
s given above. these a
k
s are a little messy to compute by hand, but easy to program. Once
again, I have had my computer graph a number of partial sums S
N
(t) =
a0
2
+

N1
k=1
a
k
cos(kt), for a specic
choice of the ramp parameter a, namely a = 0.9. The rst graph is of S
1
(t) = 1 +
a

is just a straight line

whose height is the average value of f(t). This time we get a really good likeness of the graph of f(t) before
we have included a hundred terms.
y
t
3 2
a
2 3
1
S
1
(t)
y
t
3 2
a
2 3
1
S
4
(t)
y
t
3 2
a
2 3
1
S
8
(t)
y
t
3 2
a
2 3
1
S
16
(t)
February 4, 2007 Fourier Series 9
y
t
3 2
a
2 3
1
S
32
(t)
y
t
3 2
a
2 3
1
S
64
(t)
Example 8 By a standard trig identity, the periodic function f(t) = cos
2
t obeys
f(t) = cos
2
t =
1
2
+
1
2
cos(2t) (13)
The right hand side is a Fourier cosine expansion
a0
2
+

k=1
a
k
cos(kt) with a
0
= 1, a
2
=
1
2
and all other a
k
s
zero. Just as in the complex case, the coecients in the real Fourier expansions are uniquely determined.
So the right hand side of (13) is the Fourier expansion for f(t) = cos
2
t.
Application of Fourier Series to Ordinary Dierential Equations.
Consider the RLC circuit
+

x(t)
R L
C
i(t)
+

y(t)
Were going to think of the voltage x(t) as a input signal and the voltage y(t) as an output signal. The goal
is to determine the output voltage for any given input voltage. If i(t) is the current owing at time t in the
loop as shown and q(t) is the charge on the capacitor, then the voltage across R, L and C, respectively, at
time t are Ri(t), L
di
dt
(t) and y(t) =
q(t)
C
. By Kirchhos law, these three voltages must add up to x(t) so that
Ri(t) + L
di
dt
(t) +
q(t)
C
= x(t) (14)
This is one equation in the two unknowns, i(t) and q(t). Fortunately, there is a relationship between the
two. Namely
i(t) =
dq
dt
(t) = Cy

(t) (15)
This just says that the capacitor cannot create or destroy charge on its own. All charging of the capacitor
must come from the current. Subbing (15) into (14) gives
LCy

(t) + RCy

(t) + y(t) = x(t) (16)

February 4, 2007 Fourier Series 10
We are going to consider the case in which x(t) and y(t) are both
(1)
periodic
(2)
, say with period 2. Hence
both x(t) and y(t) have Fourier series expansions of the form (9). Electrical engineers routinely use complex
Fourier series (8) in place of (9) and we shall do so too. The reason is that the computations tend to be
much simpler. Of course, in the real world x(t) and y(t) are real. When electrical engineers write
x(t) =

k=
x
k
e
ik

t
y(t) =

k=
y
k
e
ik

t
(17)
they really mean that x(t) is the real part of

k=
x
k
e
ik

t
and y(t) is the real part of

k=
y
k
e
ik

t
.
Subbing (17) into (16) gives

k=
LC
k
2

2
y
k
e
ik

t
+

k=
iRC
k

y
k
e
ik

t
+

k=
y
k
e
ik

t
=

k=
x
k
e
ik

t
which is equivalent to

k=
_
LC
k
2

2
+ iRC
k

+ 1
_
y
k
e
ik

t
=

k=
x
k
e
ik

t
The left hand side is a Fourier series expansion with coecient c
k
=
_
LC
k
2

2
+iRC
k

+1
_
y
k
. The right
hand side is a Fourier series expansion for the same function with coecient c
k
= x
k
. Because the Fourier
coecients of any function are uniquely determined (we saw that in the answer to Question #2), the k
th
Fourier coecients of the two series must be the same so that
_
LC
k
2

2
+ iRC
k

+ 1
_
y
k
= x
k
To save writing, lets call the k
th
frequency k

=
k
and lets also write a
k
= 1 LC
2
k
and b
k
= RC
k
so
that
_
LC
k
2

2
+ iRC
k

+ 1
_
= a
k
+ ib
k
. With this notation
y
k
=
1
a
k
+ib
k
x
k
We conclude that the circuit acts independently on each frequency,
k
, component of the signal. An
input signal e
i
k
t
results in an output signal
1
a
k
+ib
k
e
i
k
t
. So a real input signal Re
_
e
i
k
t
_
= cos(
k
t) results
in a real output signal
Re
_
1
a
k
+ib
k
e
i
k
t
_
= Re
_
1
a
k
+ib
k
a
k
ib
k
a
k
ib
k
e
i
k
t
_
= Re
_
a
k
ib
k
a
2
k
+b
2
k
_
cos(
k
t) + i sin(
k
t)

_
=
a
k
a
2
k
+b
2
k
cos(
k
t) +
b
k
a
2
k
+b
2
k
sin(
k
t)
because the two other terms, ia
k
sin(
k
t) ib
k
cos(
k
t), in the product (a
k
ib
k
)
_
cos(
k
t) + i sin(
k
t)

are pure imaginary. There is another, easier, way to compute the real part of
1
a
k
+ib
k
e
i
k
t
that also gives
a more illuminating form for the answer. First use polar coordinates to write a
k
+ ib
k
= r
k
e
i
k
, with
r
k
=
_
a
2
k
+ b
2
k
=
_
(1 LC
2
k
)
2
+ R
2
C
2

2
k
and
k
= tan
1 b
k
a
k
= tan
1 RC
k
1LC
2
k
. Then
Re
_
1
a
k
+ib
k
e
i
k
t
_
= Re
_
1
r
k
e
i
k
e
i
k
t
_
= Re
_
1
r
k
e
i
k
e
i
k
t
_
= Re
_
1
r
k
e
i(
k
t
k
)
_
=
1
r
k
cos(
k
t
k
)
(1)
In practice signals are applied to circuits only for nite time intervals. When the input is rst turned on, the output
usually contains some non-periodic part too. But this non-periodic part is transient meaning that it damps out. We
are implicitly assuming that all transients have essentially disappeared from the output signal.
(2)
Periodicity is really a bit of a red herring here. A similar procedure, using the Fourier transform (that we will learn about
a little later) in place of Fourier series handles non-periodic inputs.
February 4, 2007 Fourier Series 11
So the RLC cricuit has two eects on the frequency
k
part of the input signal. Its amplitude is
multiplied by
1
r
k
=
1

(1LC
2
k
)
2
+R
2
C
2

2
k
and it also undergoes a phase shift
k
= tan
1 RC
k
1LC
2
k
. Here is a
graph of A =
1

(1LC
2
)
2
+R
2
C
2

2
against

A
Note that there is a small range of frequencies that give a large amplitude response. This is the phenomenon
of resonance. That is why this circuit is often used as a lter to extract a specic frequency from the input
signal.
Power and Parsevals Relation
The energy carried by a signal f(t) is
_

|f(t)|
2
dt. For a (nonzero) periodic signal this is always
innite. So for a periodic signal the average power (energy per unit time) is a much more useful quantity. If
f(t) has period 2, it has average power
P =
1
2
_

|f(t)|
2
dt
We can express this power in terms of the Fourier coecients of f just by substituting f(t) =

k=
c
k
e
ik

t
P =
1
2
_

f(t)f(t) dt =
1
2

k,m=
_

c
k
e
ik

t
c
m
e
im

t
dt =
1
2

k,m=
c
k
c
m
_

e
i(km)

t
dt
By the orthogonality relation (2), with 17 replaced by m,
_

e
i(km)

t
dt =
_
2 if k = m
0 otherwise
So all of the terms in the double sum with k = m are zero and we are left with
P =
1
2
_

|f(t)|
2
dt =
1
2

k,m=
k=m
c
k
c
m
2 =

k=
|c
k
|
2
(18)
This is called Parsevals relation. To convert it into a statement about the Fourier coecients a
k
, b
k
, we
just need observe, from (10), that c
0
=
1
2
a
0
and, for k > 0, c
k
=
1
2
(a
k
ib
k
), c
k
=
1
2
(a
k
+ ib
k
). Assuming
that a
k
and b
k
are real, |c
0
|
2
=
1
4
|a
0
|
2
and, for k > 0, |c
k
|
2
= |c
k
|
2
=
1
4
[a
2
k
+ b
2
k
] so that
P =
1
2
_

|f(t)|
2
dt =
1
4
a
2
0
+
1
2

k=1
_
a
2
k
+ b
2
k

Suppose that we have some complicated signal f(t) =

k=
c
k
e
ik

t
and we wish to approximate it by
a simple signal that only contains the nite number of frequencies k

with, for example, k = 2, 1, 0, 1, 2.

February 4, 2007 Fourier Series 12
That is, our approximating signal is to be of the form F(t) = d
2
e
2i

t
+d
1
e
i

t
+d
0
+d
1
e
i

t
+d
2
e
2i

t
.
What should we choose for the coecients d
2
, d
1
, d
0
, d
1
and d
2
? To ensure the the error carries minimum
average power, we should choose the coecients so that
1
2
_

|f(t) F(t)|
2
dt
is as small as possible. Since
f(t) F(t) =
_

k=
c
k
e
ik

t
_

_
2

k=2
d
k
e
ik

t
_
=

k=
e
k
e
ik

t
with
e
k
=
_
c
k
d
k
if |k| 2
c
k
otherwise
Parsevals relation (18) gives that the average power carried by the error is
1
2
_

|f(t) F(t)|
2
dt =

k=
|e
k
|
2
=

2k2
|c
k
d
k
|
2
+

<k<
|k|>2
|c
k
|
2
This is minimised by choosing d
k
= c
k
for all 2 k 2.
Example 9 By Example 2, the function
f(t) =
_
1 if 0 < t <
1 if < t < 0
_
f(t) = f(t + 2)
has Fourier coecients
c
k
=
_

2
k
i if k is odd
0 if k is even
So Parsevals relation (18) gives that
1
2
_

|f(t)|
2
dt =

<k<
k odd

i
2
k

2
=

<k<
k odd
4
k
2

2
=

0<k<
k odd
8
k
2

2
Since
1
2
_

|f(t)|
2
dt =
1
2
_

1 dt = 1, we conclude that

0<k<
k odd
8
k
2

2
= 1 or
1
1
2
+
1
3
2
+
1
5
2
+ =

2
8
February 4, 2007 Fourier Series 13