13.1 Why Fourier Series?

WHY FOURIER SERIES?
THE FOURIER
SERIES OF A FUNCTION SINE AND
CHAPTER 13 C O S I N E S E R I E S I N T E G R AT I O N A N D
D I F F E R E N T I AT I O N
Fourier Series
In 1807, Joseph Fourier submitted a paper to the French Academy of Sciences in competition for
a prize offered for the best mathematical treatment of heat conduction. In the course of this work
Fourier shocked his contemporaries by asserting that “arbitrary” functions (such as might specify
initial temperatures) could be expanded in series of sines and cosines. Consequences of Fourier’s
work have had an enormous impact on such diverse areas as engineering, music, medicine, and
the analysis of data.
13.1 Why Fourier Series?
A Fourier series is a representation of a function as a series of constant multiples of sine and/or

cosine functions of different frequencies. To see how such a series might arise, we will look at a
problem of the type that concerned Fourier.
Consider a thin homogeneous bar of metal of length π , constant density and uniform cross
section. Let u (x , t ) be the temperature in the bar at time t in the cross section at x . Then (see
Section 12.8.2) u satisfies the heat equation
∂u ∂ 2u
=k 2
∂t ∂x
for 0 < x < π and t > 0. Here k is a constant depending on the material of the bar. If the left and
right ends are kept at temperature zero, then
u (0, t ) = u (π, t ) = 0 for t > 0.
These are the boundary conditions. Further, assume that the initial temperature has been
specified, say
u (x , 0) = f (x ) = x (π − x ).
This is the initial condition.
427
Copyright 2010 Cengage Learning. All Rights Reserved. May not be copied, scanned, or duplicated, in whole or in part. Due to electronic rights, some third party content may be suppressed from the eBook and/or eChapter(s).
Editorial review has deemed that any suppressed content does not materially affect the overall learning experience. Cengage Learning reserves the right to remove additional content at any time if subsequent rights restrictions require it.
428 CHAPTER 13 Fourier Series
Fourier found that the functions

2
u n (x , t ) = bn sin(nx )e−kn t
satisfy the heat equation and the boundary conditions, for every positive integer n and any number
bn . However, there is no choice of n and bn for which this function satisfies the initial condition,
which would require that
u n (x , 0) = bn sin(nx ) = x (π − x ).
for 0 ≤ x ≤ π .
We could try a finite sum of these functions, attempting a solution
N
2
u (x , t ) = bn sin(nx )e−kn t .
X
n =1
But this would require that N and numbers b1 , · · · , b N be found so that

N
u (x , 0) = x (π − x ) = bn sin(nx )
X
n =1
for 0 ≤ x ≤ π . Again, this is impossible. A finite sum of multiples of sine functions is not a
polynomial.
Fourier’s brilliant insight was to attempt an infinite superposition,
∞
2
u (x , t ) = bn sin(nx )e−kn t .
X
n =1
This function will still satisfy the heat equation and the boundary conditions u (x , 0) = u (π, 0) =
0. To satisfy the initial condition, the problem is to choose the numbers bn so that
∞
u (x , 0) = x (π − x ) = bn sin(nx )
X
n =1
for 0 ≤ x ≤ π . Fourier claimed not only that this could this be done, but that the right choice is
1 π 4 1 − (−1)n
Z
bn = x (π − x ) sin(nx ) dx = .
π 0
3 π n
With these coefficients, Fourier wrote the solution for the temperature function:
4X
∞
1 − (−1)n
u (x , t ) = sin(nx )e−kn2 t .
3 π n =1
n
The astonishing claim that
4X
∞
1 − (−1)n
x (π − x ) = sin(nx )
π n =1
n3
for 0 ≤ x ≤ π was too much for Fourier’s contemporaries to accept, and the absence of rigorous
proofs in his paper led the Academy to reject its publication (although they awarded him the
prize). However, the implications of Fourier’s work were not lost on natural philosophers of
his time. If Fourier was right, then many functions would have expansions as infinite series of
trigonometric functions.
Although Fourier did not have the means to supply the rigor his colleagues demanded, this
was provided throughout the ensuing century and Fourier’s ideas are now seen in many important
applications. We will use them to solve partial differential equations, beginning in Chapter 16.
This and the next two chapters develop the requisite ideas from Fourier analysis.
13.2 The Fourier Series of a Function 429
SECTION 13.1 PROBLEMS
1. Let 2. Let p(x ) be a polynomial. Prove that there is no number

4 N
1 − (−1)n k such that p(x ) = k sin(nx ) on [0, π ] for any positive
SN (x ) = sin(nx ). integer n .
X
π n =1
n3
3. Let pP(x ) be a polynomial. Prove that there is no finite
Construct graphs of SN (x ) and x (π − x ), for 0 ≤ x ≤ π , sum nN=1 bn sin(nx ) that is equal to p(x ) for 0 ≤ x ≤ π ,
for N = 2 and then N = 10. This will give some sense of for any choice of the numbers b1 , · · · , b N .
the correctness of Fourier’s claim that this polynomial
could be exactly represented by the infinite series
4X ∞
1 − (−1)n
sin(nx )
π3
n =1
n
on [0, π ].
13.2 The Fourier Series of a Function
Let f (x ) be defined on [− L , L ]. We want to choose numbers a0 , a1 , a2 · · · and b1 , b2 , · · ·

so that
1 ∞
f (x ) = a0 + [ak cos(k π x / L ) + bk sin(k π x / L )]. (13.1)

X
2 k =1
This is a decomposition of the function into a sum of terms, each representing the influence of a
different fundamental frequency on the behavior of the function.
To determine a0 , integrate equation (13.1) term by term to get
1 L
Z L Z
f (x )dx = a dx
−L 2 −L 0
∞ Z L Z L
ak cos(k π x / L ) dx + bk sin(k π x / L ) dx
X
+
k =1 −L −L
1
a (2 L ) = π a 0 . =
2 0
because all of the integrals in the summation are zero. Then
1 L
Z
a0 = f (x ) dx . (13.2)
L −L
To solve for the other coefficients in the proposed equation (13.1), we will use the following three
facts, which follow by routine integrations. Let m and n be integers. Then
Z L
cos(n π x / L ) sin(m π x / L ) dx = 0. (13.3)
−L
Furthermore, if n 6= m , then
Z L Z L
cos(n π x / L ) cos(m π x / L ) dx = sin(n π x / L ) sin(m π x / L ) dx = 0. (13.4)
−L −L
And, if n 6= 0, then
Z L Z L
cos (n π x / L ) dx =
2
sin2 (n π x / L ) dx = L . (13.5)
−L −L
Now let n be any positive integer. To solve for an , multiply equation (13.1) by cos(n π x / L ) and
integrate the resulting equation to get
1
Z L Z L
f (x ) cos(n π x / L ) dx = a0 cos(n π x / L ) dx
−L 2 −L
∞ Z L Z L
ak cos(k π x / L ) cos(n π x / L ) dx + bk sin(k π x / L ) cos(n π x / L ) dx
X
+ .
k =1 −L −L
Because of equations (13.3) and (13.4), all of the terms on the right are zero except the coefficient
of an , which occurs in the summation when k = n . The last equation reduces to
Z L Z L
f (x ) cos(n π x / L ) dx = an cos2 (n π x / L ) dx = an L
−L −L
by equation (13.5). Therefore

1 L
Z
an = f (x ) cos(n π x / L ) dx . (13.6)
L L
This expression contains a0 if we let n = 0.
Similarly, if we multiply equation (13.1) by sin(n π x / L ) instead of cos(n π x / L ) and
integrate, we obtain
1 L
Z
bn = f (x ) sin(n π x / L ) dx . (13.7)
L −L
The numbers
1 L
Z
an = f (x ) cos(n π x / L ) dx for n = 0, 1, 2, · · · (13.8)
L −L
1 L
Z
f (x ) sin(n π x / L ) dx for n = 1, 2, · · ·
bn = (13.9)
L −L
are called the Fourier coefficients of f on [ L , L ]. When these numbers are used, the series
(13.1) is called the Fourier series of f on [ L , L ].
EXAMPLE 13.1
Let f (x ) = x − x 2 for −π ≤ x ≤ π . Here L = π . Compute

1 π 2
Z
a0 = (x − x 2 ) dx = − π 2 ,
π −π 3
1 π
Z
an = (x − x 2 ) cos(nx )dx
π −π
4 sin(n π ) − 4n π cos(n π ) − 2n 2 π 2 sin(n π )

=
π n3
4 4
=− cos(n π ) = − (−1)n
n2 n2
4(−1)n+1
= ,
n2
and
1 π
Z
bn = (x − x 2 ) sin(nx ) dx
π −π
2 sin(n π ) − 2n π cos(n π )
=
π n2
2 2
=− cos(n π ) = − (−1)n
n n
2(−1)n+1
= .
n
We have used the facts that sin(n π ) = 0 and cos(n π ) = (−1)n if n is an integer.
The Fourier series of f (x ) = x − x 2 on [−π, π] is
1 ∞
4(−1)n+1 2(−1)n+1

− π2 + cos nx sin nx
X
( ) + ( ) .
3 n =1
n 2 n
This example illustrates a fundamental issue. We do not know what this Fourier series con-
verges to. We need something that establishes a relationship between the function and its Fourier
series on an interval. This will require some assumptions about the function.
Recall that f is piecewise continuous on [a , b] if f is continuous at all but perhaps finitely
many points of this interval, and, at a point where the function is not continuous, f has finite
limits at the point from within the interval. Such a function has at worst jump discontinuities, or
finite gaps in the graph, at finitely many points. Figure 13.1 shows a typical piecewise continuous
function.
If a < x0 < b, denote the left limit of f (x ) at x0 as f (x0 −), and the right limit of f (x ) at x0
as f (x0 +):
f (x0 −) = hlim
→0+
f (x0 − h ) and f (x0 +) = hlim
→0+
f (x0 + h ).
If f is continuous at x0 , then these left and right limits both equal f (x0 ).
FIGURE 13.1 A piecewise con-

tinuous function.
1
x
–4 –2 0 0 2 4 6
–1
–2
–3
FIGURE 13.2 f in Example 13.2.
EXAMPLE 13.2
Let
x for −3 ≤ x < 2,
(
f (x ) =
1/x for 2 ≤ x ≤ 4.
f is piecewise continuous on [−3, 4], having a single discontinuity at x = 2. Furthermore,
f (2−) = 2 and f (2+) = 1/2. A graph of f is shown in Figure 13.2.
f is piecewise smooth on [a , b] if f is piecewise continuous and f ′ exists and is continuous

at all but perhaps finitely many points of (a , b).
EXAMPLE 13.3
The function f of Example 13.2 is differentiable on (−3, 4) except at x = 2:

1 for −3 < x < 2
(
f ′ (x ) =
−1 /x 2 for 2 < x < 4.
This derivative is itself piecewise continuous. Therefore f is piecewise smooth on [−3, 4].
We can now state a convergence theorem.
THEOREM 13.1 C o nve r g e n c e o f Fo u r i e r S e r i e s
Let f be piecewise smooth on [− L , L ]. Then, for each x in (− L , L ), the Fourier series of f on

[− L , L ] converges to
1
( f (x +) + f (x −)).
2
FIGURE 13.3 Convergence of a Fourier

series at a jump discontinuity.
At both − L and L this Fourier series converges to

1
( f ( L −) + f (− L +)).
2
At any point in (− L , L ) at which f (x ) is continuous, the Fourier series converges to f (x ),
because then the right and left limits at x are both equal to f (x ). At a point interior to the
interval where f has a jump discontinuity, the Fourier series converges to the average of the left
and right limits there. This is the point midway in the gap of the graph at the jump discontinuity
(Figure 13.3). The Fourier series has the same sum at both ends of the interval.
EXAMPLE 13.4
Let f (x ) = x − x 2 for −π ≤ x ≤ π . In Example 13.1 we found the Fourier series of f on [−π, π].
Now we can examine the relationship between this series and f (x ).
f ′ (x ) = 1 − 2x is continuous for all x , hence f is piecewise smooth on [−π, π]. For −π <
x < π , the Fourier series converges to x − x 2 . At both π and −π , the Fourier series converges to
1 1
( f (π−) + f (−π+)) = ((π − π 2 ) + (−π − (−π )2 ))
2 2
1
= (−2π 2 ) = −π 2 .
2
Figures 13.4, 13.5, and 13.6 show the fifth, tenth and twentieth partial sums of this Fourier
series, together with a graph of f for comparison. The partial sums are seen to approach the
function as more terms are included.
EXAMPLE 13.5
Let f (x ) = ex . The Fourier coefficients of f on [−1, 1] are

Z 1
a0 = ex dx = e − e−1 ,
−1
1
(e − e−1 )(−1)n
Z
an = ex cos(n π x ) dx = ,
−1 1 + n2π 2
x x
–3 –2 –1 0 1 2 3 –3 –2 –1 0 1 2 3
0 0
–2 –2
–4 –4
–6 –6
–8 –8
–10 –10
–12 –12
FIGURE 13.4 Fifth partial sum of the FIGURE 13.5 Tenth partial sum in Example
Fourier series in Example 13.4. 13.4.
x
–3 –2 –1 0 1 2 3
–2
–4
–6
–8
–10
–12
FIGURE 13.6 Twentieth partial sum in

Example 13.4.
and
1
(e − e−1 )(−1)n (n π )
Z
bn = ex sin(n π x ) dx = − .
−1 1 + n2π 2
The Fourier series of e on [−1, 1] is x
1 ∞
(−1)n

(e − e−1 ) + (e − e−1 ) (cos(n π x ) − n π sin(n π x )).
X
2 n =1
1 + n2π 2
Because ex is continuous with a continuous derivative for all x , this series converges to
ex for −1 < x < 1
(
2 (e + e )
1 −1
for x = 1 and for x = −1.
2.5
1.5
0.5
–1 –0.5 0 0.5 1
x
FIGURE 13.7 Thirtieth partial sum of the

Fourier series in Example 13.5.
Figure 13.7 shows the thirtieth partial sum of this series, suggesting its convergence to the
function except at the endpoints −1 and 1.
EXAMPLE 13.6
Let
5 sin(x ) for −2π ≤ x < −π/2




4 for x = −π/2




f (x ) = x 2 for −π/2 < x < 2
8 cos(x ) for 2 ≤ x < π




4x for π ≤ x ≤ 2π .



f is piecewise smooth on [−2π, 2π]. The Fourier series of f on [−2π, 2π ] converges to:
5 sin
(x ) for −2π < x < −π/2


 1 π2
2 4 −5 for x = −π/2





x2 for −π/2 < x < 2





2 (4 + 8 cos(2)) for x = 2
1


8 cos(x ) for 2 < x < π
 1 (4π − 8) for x = π



 2
4 x for −π < x < 2π





4π for x = 2π and x = −2π .


This conclusion does not require that we write the Fourier series.
13.2.1 Even and Odd Functions
A function f is even on [− L , L ] if its graph on [− L , 0] is the reflection across the vertical

axis of the graph on [0, L ]. This happens when f (−x ) = f (x ) for 0 < x ≤ L . For example,
x 2n and cos(n π x / L ) are even on any [− L , L ] for any positive integer n .
A function f is odd on [− L , L ] if its graph on [− L , 0) is the reflection through the
origin of the graph on (0, L ]. This means that f is odd when f (−x ) = − f (x ) for 0 < x ≤ L .
For example, x 2n+1 and sin(n π x / L ) are odd on [− L , L ] for any positive integer n .
Figures 13.8 and 13.9 show typical even and odd functions, respectively.
A product of two even functions is even, a product of two odd functions is even, and a
product of an odd function with an even function is odd.
If f is even on [− L , L ] then
Z L Z L
f (x ) dx = 2 f (x ) dx
−L 0
and if f is odd on [− L , L ] then

Z L
f (x ) dx = 0.
−L
These facts are sometimes useful in computing Fourier coefficients. If f is even, only the cosine
terms and possibly the constant term will appear in the Fourier series, because f (x ) sin(n π x / L )
is odd and the integrals defining the sine coefficients will be zero. If the function is odd
then f (x ) cos(n π x / L ) is odd and the Fourier series will contain only the sine terms, since
the integrals defining the constant term and the coefficients of the cosine terms will be
zero.
35
60
30
25 40
20 20
x
15 –4 –2 0 0 2 4
10
–20
5
–40
0
–6 –4 –2 0 2 4 6
x –60
FIGURE 13.8 A typical even function. FIGURE 13.9 A typical odd function.
1
x
–3 –2 –1 0 1 2 3
0
–1
–2
–3
FIGURE 13.10 Twentieth partial sum of the

series in Example 13.7.
EXAMPLE 13.7
We will compute the Fourier series of f (x ) = x on [−π, π].
Because x cos(nx ) is an odd function on [−π, π] for n = 0, 1, 2, · · · , each an = 0. We need
only compute the bn ′ s :
1 π
Z
bn = x sin(nx ) dx
π −π
2 π
Z
= x sin(nx ) dx
π 0
2 2x
π
= sin(nx ) − cos(nx )
n2π nπ 0
2 2
=− cos(n π ) = (−1)n+1 .
n n
The Fourier series of x on [−π, π] is
∞
2
(−1)n+1 sin(nx ).
X
n =1
n
This converges to x for −π < x < π , and to 0 at x = ±π . Figure 13.10 shows the twentieth
partial sum of this Fourier series compared to the function.
EXAMPLE 13.8
We will write the Fourier series of x 4 on [−1, 1]. Since x 4 sin(n π x ) is an odd function on [−1, 1]
for n = 1, 2, · · · , each bn = 0. Compute
2
Z 1
a0 = 2 x 4 dx =
0 5
and Z 1
an = 2 x 4 cos(n π x )dx
0
1
(n π x )4 sin(n π x ) + 4(n π x )3 cos(n π x )

=2
(n π )5 0
0.8
0.6
0.4
0.2
0
–1 –0.5 0 0.5 1
x
FIGURE 13.11 Tenth partial sum of the

series of Example 13.8.
1
−12(n π x )2 sin(n π x ) − 24n π x cos(n π x ) + 24 sin(n π x )

+2
(n π )5 0
n2π 2 − 6
= 8 4 4 (−1)n .
nπ
The Fourier series is
1 X∞
8(−1)n 2 2

+ (n π − 6) cos(n π x ).
5 n=1 n 4 π 4
By Theorem 13.1, this series converges to x 4 for −1 ≤ x ≤ 1. Figure 13.11 shows the
twentieth partial sum of this series, compared with the function.
13.2.2 The Gibbs Phenomenon
A.A. Michelson was a Prussian-born physicist who teamed with E.W. Morley of Case-Western
Reserve University to show that the postulated “ether,” a fluid which supposedly permeated all of
space, had no effect on the speed of light. Michelson also built a mechanical device for construct-
ing a function from its Fourier coefficients. In one test, he used eighty coefficients for the series
of f (x ) = x on [−π, π] and noticed unexpected jumps in the graph near the endpoints. At first,
he thought this was a problem with his machine. It was subsequently found that this behavior is
characteristic of the Fourier series of a function at a point of discontinuity. In the early twentieth
century, the Yale mathematician Josiah Willard Gibbs finally explained this behavior.
To illustrate the Gibbs phenomenon, expand f in a Fourier series on [−π, π], where
−π/4 for −π ≤ x < 0


f (x ) = 0 for x = 0
π/4 for 0 < x ≤ π .


This function has a jump discontinuity at 0, but its Fourier series on [−π, π] converges at 0 to
1 1 π π
( f (0+) + f (0−)) = − = 0 = f (0).
2 2 4 4
0.8
0.4
0
–3 –2 –1 0 1 2 3
x
–0.4
–0.8
FIGURE 13.12 The Gibbs phenomenon.
The Fourier series therefore converges to the function on (−π, π ). This series is
∞
1
sin((2n − 1)x ).
X
n =1
2n − 1
Figure 13.12 shows the fifth and twenty-fifth partial sums of this series, compared to the
function. Notice that both of these partial sums show a peak near 0, the point of discontinuity of
f . Since the partial sums SN approach the function as N → ∞, we might expect these peaks to
flatten out, but they do not. Instead they remain roughly the same height, but move closer to the
y -axis as N increases. This is the Gibbs phenomenon.
Postscript We will add two comments on the ideas of this section, the first practical and the
second offering a broader perspective.
1. Writing and graphing partial sums of Fourier series are computation intensive activities.
Evaluating integrals for the coefficients is most efficiently done in MAPLE using the
int command, and partial sums are easily graphed using the sum command to enter the
partial sum and then the plot command for the graph.
2. Partial sums of Fourier series can be viewed from the perspective of orthogonal pro-
jections onto a subspace of a vector space (Sections 6.6 and 6.7). Let PC [− L , L ] be
the vector space of functions that are piecewise continuous on [− L , L ] and let S be the
subspace spanned by the functions
C0 (x ) = 1, Cn (x ) = cos(n π x / L ) and Sn (x ) = sin(n π x / L ) for n = 1, 2, · · · , N .
A dot product can be defined on P L [− L , L ] by
Z L
f ·g= f (x )g (x ) dx .
−L
Using this dot product, these functions form an orthogonal basis for S . If f if piecewise
continuous on [− L , L ], the orthogonal projection of f onto S is
N
f · C0 f · Cn f · Sn

fS = C + C + S .
X
C0 · C0 0 n=1 Cn · Cn n Sn · Sn n
Compare these coefficients in the orthogonal projection with the Fourier coefficients of f on
[− L , L ]. First
RL
f · C0 −L
f (x ) dx
=
C0 · C0
RL
−L
dx
1 L
1
Z
= f (x ) dx = a0 .
2L −L 2
Next,
L
f · Cn f (x ) cos(n π x / L ) dx
R
−L
= RL
Cn · Cn −L
cos2 (n π x / L ) dx
1 L
Z
= f (x ) cos(n π x / L ) dx = an ,
L −L
and similarly,
f · Sn 1 L
Z
= f (x ) sin(n π x / L ) dx = bn .
Sn · Sn L −L
Thus, the orthogonal projection f S of f onto S is exactly the N th partial sum of the Fourier series
of f on [− L , L ].
This broader perspective of Fourier series will provide a unifying theme when we consider
general eigenfunction expansions in Chapter 15.
In each of Problems 1 through 12, write the Fourier series 10. f (x ) = cos(x /2) − sin(x ), −π ≤ x ≤ π
of the function on the interval and determine the sum of
the Fourier series. Graph some partial sums of the series, 11. f (x ) = cos(x ), −3 ≤ x ≤ 3
compared with the graph of the function. (
1 − x for −1 ≤ x ≤ 0
12. f (x ) =
1. f (x ) = 4, −3 ≤ x ≤ 3 0 for 0 < x ≤ 1
2. f ( x ) = − x , −1 ≤ x ≤ 1
In each of Problems 13 through 19, use the convergence
3. f (x ) = cosh(π x ), −1 ≤ x ≤ 1 theorem to determine the sum of the Fourier series of the
4. f (x ) = 1 − |x |, −2 ≤ x ≤ 2 function on the interval. It is not necessary to write the
series to do this.
for −π ≤ x ≤ 0
−4
(
5. f (x ) =
4 for 0 < x ≤ π 2x for −3 ≤ x < −2



6. f (x ) = sin(2x ), −π ≤ x ≤ π 13. f (x ) = 0 for −2 ≤ x < 1
x 2 for 1 ≤ x ≤ 3

7. f (x ) = x 2 − x + 3, −2 ≤ x ≤ 2

−x for −5 ≤ x < 0 2x − 2 for −π ≤ x ≤ 1

( (
8. f (x ) = 14. f (x ) = 3 for 1 < x ≤ −π
1 + x for 0 ≤ x ≤ 5
2
1 for −π ≤ x < 0 x 2 for −π ≤ x ≤ 0

( (
9. f (x ) = 15. f (x ) =
2 for 0 ≤ x ≤ π 2 for 0 < x ≤ π
13.3 Sine and Cosine Series 441
cos(x ) for −2 ≤ x < 0 −2 for −4 ≤ x ≤ −2

 
16. f (x ) = sin(x )

 for 0 ≤ x ≤ 2 19. f (x ) = 1 + x 2for −2 < x ≤ 2
0 for 2 < x ≤ 4


−1 for −4 ≤ x < 0
(
17. f (x ) = 20. Using Problem 14, write the Fourier series of the
1 for 0 ≤ x ≤ 4 function and plot some partial sums, pointing out the
occurrence of the Gibbs phenomenon at the points of
for −1 ≤ x < 1/2
0 discontinuity of the function.


18. f (x ) = 1 for 1/2 ≤ x ≤ 3/4 21. Carry out the program of Problem 20 for the function
2 for 3/4 < x ≤ 1 of Problem 16.


13.3 Sine and Cosine Series

If f is piecewise continuous on [− L , L ], we can represent f (x ) at all but possibly finitely many
points of [− L , L ] by its Fourier series. This series may contain just sine terms, just cosine terms,
or both sine and cosine terms. We have no control over this.
If f is defined on the half interval [0, L ], we can write a Fourier cosine series (containing
just cosine terms) and a Fourier sine series (containing just sine terms) for f on [0, L ].
13.3.1 Cosine Series
Suppose f (x ) is defined for 0 ≤ x ≤ L . To get a pure cosine series on this interval, imagine
reflecting the graph of f across the vertical axis to obtain an even function g defined on [− L , L ]
(see Figure 13.13).
Because g is even, its Fourier series on [− L , L ] has only cosine terms and perhaps the
constant term. But g (x ) = f (x ) for 0 ≤ x ≤ L , so this gives a cosine series for f on [0, L ].
Furthermore, because g is even, the coefficients in the Fourier series of g on [− L , L ] are
1 L
Z
an = g (x ) cos(n π x / L ) dx
L −L
2 L
Z
= g (x ) cos(n π x / L ) dx
L 0
2 L
Z
= f (x ) cos(n π x / L ) dx .
L 0
FIGURE 13.13 Even extension of a

function defined on [0, L ].
for n = 0, 1, 2, · · · . Notice that we can compute an strictly in terms of f on [0, L ]. The construc-
tion of g showed us how to obtain this cosine series for f , but we do not need g to compute the
coefficients of this series.
Based on these ideas, define the Fourier cosine coefficients of f on [0, L ] to be the numbers
2 L
Z
an = f (x ) cos(n π x / L ) dx (13.10)
L 0
for n = 0, 1, 2, · · · . The Fourier cosine series for f on [0, L ] is the series
1 ∞
a0 + an cos(n π x / L ) (13.11)
X
2 n =1
in which the an s are the Fourier cosine coefficients of f on [0, L ].

′
By applying Theorem 13.1 to g , we obtain the following convergence theorem for cosine
series on [0, L ].
THEOREM 13.2 C o nve r g e n c e o f Fo u r i e r C o s i n e S e r i e s
Let f be piecewise smooth on [0, L ]. Then

1. If 0 < x < L , the Fourier cosine series for f on [0, L ] converges to
1
( f (x +) + f (x −)).
2
2. At 0 this cosine series converges to f (0+).
3. At L this cosine series converges to f ( L −).
EXAMPLE 13.9
Let f (x ) = e2x for 0 ≤ x ≤ 1. We will write the cosine expansion of f on [0, 1]. The coefficients
are
Z 1
a0 = 2 e2x dx = e2 − 1
0
and for n = 1, 2, · · · ,
Z 1
an = 2 e2x cos(n π x ) dx
0
1
4e2x cos(n π x ) + 2n π e2x sin(n π x )

=
4 + n2π 2 0
e2 (−1)n − 1
=4 .
4 + n2π 2
13.3 Sine and Cosine Series 443
The cosine series for e2x on [0, 1] is

1 2 ∞
e2 (−1)n − 1
(e − 1) + 4 cos(n π x ).
X
2 n =1
4 + n 2π 2
This series converges to

e
2x for 0 < x < 1


1 for x = 0
e2 for x = 1.


This Fourier cosine series converges to e2x for 0 ≤ x ≤ 1. Figure 13.14 shows e2x and the fifth
partial sum of this cosine series.
13.3.2 Sine Series
We can also write an expansion of f on [0, L ] that contains only sine terms. Now reflect the
graph of f on [0, L ] through the origin to create an odd function h on [− L , L ], with h (x ) = f (x )
for 0 < x ≤ L (Figure 13.15).
1
0 0.2 0.4 0.6 0.8 1
x
FIGURE 13.14 Fifth partial sum of the

cosine series in Example 13.9.
y
FIGURE 13.15 Odd extension of a func-

tion defined on [0, L ].
The Fourier expansion of h on [− L , L ] has only sine terms because h is odd on [− L , L ]. But
h (x ) = f (x ) on [0, L ], so this gives a sine expansion of f on [0, L ]. This suggests the following
definitions.
The Fourier sine coefficients of f on [0, L ] are

2 L
Z
bn = f (x ) sin(n π x / L ) dx . (13.12)
L 0
for n = 1, 2, · · · . With these coefficients, the series
∞
bn sin(n π x / L ) (13.13)
X
n =1
is the Fourier sine series for f on [0, L ].
Again, we have a convergence theorem for sine series directly from the convergence theorem
for Fourier series.
THEOREM 13.3 C o nve r g e n c e o f Fo u r i e r S i n e S e r i e s
Let f be piecewise smooth on [0, L ]. Then

1. If 0 < x < L , the Fourier sine series for f on [0, L ] converges to
1
( f (x +) + f (x −)).
2
2. At x = 0 and x = L , this sine series converges to 0.
Condition (2) is obvious because each sine term in the series vanishes at x = 0 and at x = L ,
regardless of the values of the function there.
EXAMPLE 13.10
Let f (x ) = e2x for 0 ≤ x ≤ 1. We will write the Fourier sine series of f on [0, 1]. The coefficients
are
Z 1
bn = 2 e2x sin(n π x ) dx
0
1
−2n π e2x cos(n π x ) + 4e2x sin(n π x )

=
4 + n2π 2 0
n π(1 − (−1) e ) n 2
=2 .
4 + n2π 2
The sine expansion of e2x on [0, 1] is
∞
n π(1 − (−1)n e2 )
2 sin(n π x ).
X
n =1
4 + n2π 2
This series converges to for 0 < x < 1 and to 0 at x = 0 and x = 1. Figure 13.16 shows
e2x
the function and the fortieth partial sum of this sine expansion.
13.4 Integration and Differentiation of Fourier Series 445
0
0 0.2 0.4 0.6 0.8 1
x
FIGURE 13.16 Fortieth partial sum of the sine

series of Example 13.10.

In each of Problems 1 through 10, write the Fourier cosine 1 for 0 ≤ x < 1

and sine series for f on the interval. Determine the sum

8. f (x ) = 0 for 1 ≤ x ≤ 3
of each series. Graph some of the partial sums of these
−1 for 3 < x ≤ 5

series.

1. f (x ) = 4, 0 ≤ x ≤ 3 x 2 for 0 ≤ x < 1
(
9. f (x ) =
(
1 for 0 ≤ x ≤ 1 1 for 1 ≤ x ≤ 4
2. f (x ) =
−1 for 1 < x ≤ 2 10. f (x ) = 1 − x 3 , 0 ≤ x ≤ 2
0 for 0 ≤ x ≤ π 11. Sum the series ∞n=1 (−1)n /(4n 2 − 1). Hint: Expand
( P
3. f (x ) = sin(x ) in a cosine series on [0, π ] and choose an
cos(x ) for π < x ≤ 2π
appropriate value of x .
4. f (x ) = 2x , 0 ≤ x ≤ 1
12. Let f (x ) be defined on [− L , L ]. Prove that f can
5. f (x ) = x 2 , 0 ≤ x ≤ 2 be written as a sum of an even function and an odd
6. f ( x ) = e−x , 0 ≤ x ≤ 1 function on this interval.
x for 0 ≤ x ≤ 2
(
7. f (x ) = 13. Determine all functions on [− L , L ] that are both even
2 − x for 2 < x ≤ 3 and odd.
13.4 Integration and Differentiation of Fourier Series

Term by term differentiation of a Fourier series may lead to nonsense.
EXAMPLE 13.11
Let f (x ) = x for −π ≤ x ≤ π . The Fourier series is
∞
2
f (x ) = x = (−1)n+1 sin(nx )
X
n =1
n
for −π < x < π . Differentiate this series term by term to get

∞
2(−1)n+1 cos(nx ).
X
n =1
This series does not converge on (−π, π ).
However, under fairly mild conditions, we can integrate a Fourier series term by term.
THEOREM 13.4 Integration of Fourier Series
Let f be piecewise continuous on [− L , L ], with Fourier series

1 ∞
a0 + (an cos(n π x / L ) + bn sin(n π x / L )).

X
2 n =1
Then, for any x with − L ≤ x ≤ L ,

1
Z x
f (t ) dt = a0 (x + L )
−L 2
LX ∞
1
an sin(n π x / L ) − bn (cos(n π x / L ) − (−1)n ) .

+
π n =1
n
The expression on the right in this equation is exactly what we obtain by integrating the
Fourier series term by term, from − L to x . This means that we can integrate any piecewise
continuous function f from − L to x by integrating its Fourier series term by term. This is true
even if the function has jump discontinuities and its Fourier series does not converge to f (x ) for
all x in [− L , L ]. Notice, however, that the result of this integration is not a Fourier series.
EXAMPLE 13.12
From Example 13.11,

∞
2
f (x ) = x = (−1)n+1 sin(nx )
X
n =1
n
on (−π, π ). Term by term differentiation results in a series that does not converge on the interval.
However, we can integrate this series term by term:
1
Z x
t dt = (x 2 − π 2 )
−π 2
∞
2 x
Z
(−1) n +1
sin(nt ) dt
X
=
n =1
n −π
∞
2 1 1

(−1) n +1
cos(nx ) + cos(n π )
X
= −
n =1
n n n
∞
2
(−1)n+1 (cos(nx ) − (−1)n ).
X
=
n =1
n
Proof Define
x
1
Z
F (x ) = f (t ) dt − a0 x
−L 2
for − L ≤ x ≤ L . Then F is continuous on [− L , L ] and
1
F (− L ) = F ( L ) = a0 L .
2
Furthermore,
1
F ′ ( x ) = f ( x ) − a0
2
at every point on (− L , L ) at which f is continuous. Therefore F ′ (x ) is piecewise continuous on
[− L , L ], and the Fourier series of F (x ) converges to F (x ) on this interval. This series is
1 ∞ nπ x n π x
F ( x ) = A0 + An cos + Bn sin
X
.
2 n =1
L L
Here we obtain, using integration by parts,
1 L nπ x
Z
An = F (t ) cos dt
L −L L
1 1
L
L nπ x L L nπ x
Z
= F (t ) sin − sin F ′ (t ) dt
L nπ L −L L −L n π L
1 L
1 nπ x
Z
=− f (t ) − a0 sin dt
nπ −L 2 L
1 L nπ x 1 L nπ x
Z Z
=− f (t ) sin dt + a0 sin dt
n π −L L 2n π −L L
L
= − bn .
nπ
Similarly,
1L L nπ x
Z
Bn = an , F (t ) sin dt =
−L Ln π L
where the an ′ s and bn ′ s are the Fourier coefficients of f (x ) on [− L , L ]. Therefore the Fourier
series of F (x ) has the form
1 LX 1
∞ nπ x n π x
F ( x ) = A0 + −bn cos + an sin
2 π n=1 n L L
for − L ≤ x ≤ L . To determine A0 , write
L LX
∞
F ( L ) = a0 − bn cos(n π )
2 π n =1
1 LX ∞
1

= A0 − bn (−1)n .
2 π n=1 n
This gives us
2L X
∞
1

A0 = La0 + bn (−1)n .
π n =1
n
Upon substituting these expressions for A0 , An , and Bn into the Fourier series for F (x ), we obtain
the conclusion of the theorem.
Valid term by term differentiation of a Fourier series requires stronger conditions.
THEOREM 13.5 Differentiation of Fourier Series
Let f be continuous on [− L , L ] and suppose that f (− L ) = f ( L ). Let f ′ be piecewise

continuous on [− L , L ]. Then the Fourier series of f on [− L , L ] converges to f (x ) on [− L , L ]:
1 ∞
f (x ) = a0 + (an cos(n π x / L ) + bn sin(n π x / L ))

X
2 n =1
for − L ≤ x ≤ L . Further, at each x in (− L , L ) at which f ′′ (x ) exists, the term by term derivative

of the Fourier series converges to the derivative of the function:
∞
nπ
f ′ (x ) = (−an sin(n π x / L ) + bn cos(n π x / L )).
X
n =1
L
The idea of a proof of this theorem is to begin with the Fourier series for f ′ (x ), noting
that this series converges to f ′ (x ) at each point where f ′′ exists. Use integration by parts to
relate the Fourier coefficients of f ′ (x ) to those for f (x ), similar to the strategy used in proving
Theorem 13.4,
EXAMPLE 13.13
Let f (x ) = x 2 for −2 ≤ x ≤ 2. By the Fourier convergence theorem,
4 16 X
∞
(−1)n+1
2
x = + 2 cos(n π x /2)
3 π n=1 n 2
for −2 ≤ x ≤ 2. Only cosine terms appear in this series because x 2 is an even function. Now,
f ′ (x ) = 2x is continuous and f is twice differentiable for all x . Therefore, for −2 < x < 2,
8X
∞
(−1)n+1
f ′ (x ) = 2x = sin(n π x /2).
π n =1
n
This can be verified by expanding 2x in a Fourier series on [−2, 2].
Fourier coefficients, and Fourier sine and cosine coefficients, satisfy an important set of
inequalities called Bessel’s inequalities.
THEOREM 13.6 Bessel’s Inequalities
Suppose 0L g (x ) dx exists.
R
1. The Fourier sine coefficients bn of g (x ) on [0, L ] satisfy

∞
2 L
Z
bn2 ≤ (g (x ))2 dx .
X
n =1
L 0
2. The Fourier cosine coefficients an of g (x ) on [0, L ] satisfy

1 2 X ∞
2 L
Z
a0 + an2 ≤ (g (x ))2 dx .
2 n =1
L 0
RL
3. If −L
g (x ) dx exists, then the Fourier coefficients of f (x ) on [− L , L ] satisfy
1 2 X ∞
1 L
Z
a0 + (an2 + bn2 ) ≤ (g (x ))2 dx .
2 n =1
L −L
In particular, the sum of the squares of the coefficients in a Fourier series (or cosine or sine
series) converges.
We will prove conclusion (1). The argument is notationally simpler than that for conclusions
(2) and (3), but contains the ideas involved.
Proof of (1) The Fourier sine series of g (x ) on [0, L ] is
∞ nπ x
bn sin
X
,
n =1
L
where
2 L nπ x
Z
bn = g (x ) sin dx .
L 0 L
The N th partial sum of this sine series is
N nπ x
SN (x ) = bn sin
X
.
n =1
L
Then
Z L
0≤ (g (x ) − SN (x ))2 dx
0
Z L Z L Z L
= (g (x )) 2
dx − 2 g (x ) SN (x ) dx + ( SN (x ))2 dx
0 0 0
L L N nπ x
Z Z !
(g (x ))2 dx − 2 g (x ) bn sin dx
X
=
0 0 n =1
L
N nπ x N
L
kπ x
Z ! !
bn sin bk sin dx
X X
+
0 n =1
L k =1
L
Z L N Z L nπ x
(g (x ))2 dx − 2 bn g (x ) sin dx
X
=
0 n =1 0 L
N
N X nπ x
L
kπ x
Z
bn bk sin sin dx
X
+
n =1 k =1 0 L L
N N
L
LX
Z
(g (x ))2 dx − bn ( Lbn ) + bn bn .
X
=
0 n =1
2 n =1
Here we have used the fact that

L nπ x kπ x 0 for n 6= k ,
Z (
sin sin dx =
0 L L L /2 for n = k .
We therefore have
N N
L
LX
Z
0≤ (g (x ))2 dx − L bn2 + bn2 ,
X
0 n =1
2 n =1
and this gives us

N
2 L
Z
2
bn ≤ ( f (x ))2 dx .
X
n =1
L 0
Since this is true for all positive integers N , we can let N → ∞ to obtain
∞
2 L
Z
2 2
bn ≤ ( f (x )) dx .
X
n =1
L 0
EXAMPLE 13.14
We will use Bessel’s inequality to derive an upper bound for the sum of a series. Let f (x ) = x 2
on [−π, π]. The Fourier series is
1 ∞
(−1)n
f (x ) = π 2 + 4 2 cos(nx )
X
3 n =1
n
for −π ≤ x ≤ π . Here a0 = 2π 2 /3 and an = 4(−1)n / n 2 , while each bn = 0 (x 2 is an even function).
By Bessel’s inequality,
1 2π 2 X ∞
4(−1)n 2 1 π 4 2
Z
+ ≤ x dx = π 4 .
2 3 n =1
n 2 π −π 5
Then
∞
1 2 2 8π 4

16 π4 =
X
≤ − .
n =1
n4 5 9 45
Then
∞
X 1 π4
≤ ,
n =1
n4 90
which is approximately 1.0823. Infinite series are generally difficult to sum, so it is sometimes
useful to be able to derive an upper bound.
With stronger assumptions than just existence of the integral over the interval, we can derive
an important equality satisfied by the Fourier coefficients of a function on [− L , L ] or by the
Fourier sine or cosine coefficients of a function on [0, L ]. We will state the result for f (x )
defined on [− L , L ].
THEOREM 13.7 Parseval’s Theorem
Let f be continuous on [− L , L ] and let f ′ be piecewise continuous. Suppose that f (− L ) =

f ( L ). Then the Fourier coefficients of f on [− L , L ] satisfy
1 2 X ∞
1 L
Z
2 2
a + (a + b ) = f (x )2 dx .
2 0 n=1 n n L −L
Proof Begin with the fact that, from the Fourier convergence theorem,
1 ∞
f (x ) = a0 + (an cos(n π x / L ) + bn sin(n π x / L )).

X
2 n =1
Multiply this series by f (x ) to get
1 ∞
f ( x )2 = a 0 f ( x ) + (an f (x ) cos(n π x / L ) + bn f (x ) sin(n π x / L )).

X
2 n =1
We can integrate this equation term by term (Theorem 13.4). In doing this, observe that the
integrals in the series on the right are Fourier coefficients. This yields Parseval’s theorem.
EXAMPLE 13.15
We will apply Parseval’s theorem to sum a series. The Fourier coefficients of cos(x /2) on [−π, π]
are
1 π
4
Z
a0 = cos(x /2) dx =
π −π π
and, for n = 1, 2, · · · ,,
1 π
4 (−1)n
Z
an = cos(x /2) cos(nx ) dx = − .
π −π π 4n 2 − 1
Each bn = 0 because cos(x /2) is an even function. By Parseval’s theorem,

2
1 4 ∞
4 (−1)n 2 1 π 2
Z
cos (x /2) dx = 1.
X
+ =
2 π n =1
π 4 n 2 −1 π −π
After some routine manipulation, this yields

∞
X 1 π2 − 8
= .
n =1
(4n 2 − 1)2 16
We conclude this section with sufficient conditions for a Fourier series to converge
uniformly.
THEOREM 13.8
Let f be continuous on [− L , L ], and let f ′ be piecewise continuous. Suppose f ( L ) = f (− L ).

Then the Fourier series for f (x ) on [− L , L ] converges absolutely and uniformly to f (x ) on
[− L , L ].
A proof is outlined in Problem 6.

1. Let 5. Let f and f ′ be piecewise continuous on [− L , L ]. Use
(
0 for −π ≤ x ≤ 0 Bessel’s inequality to show that
f (x ) =
x for 0 < x ≤ π . Z L nπ x
lim f (x ) cos dx
n →∞ −L L
(a) Write the Fourier series of f (x ) on [−π, π ] and Z L nπ x
show that this series converges to f (x ) on (−π, π ). = lim f (x ) sin dx = 0.
n →∞ −L L
(b) Use Theorem 12.5 to show that this series can be
integrated term by term. This result is called Riemann’s lemma.
(c) Use the results of (a) and R(b) to obtain a trigono- 6. Prove Theorem 13.8 by filling in the details of the fol-
metric series expansion of −πx f (t )dt on [−π, π ]. lowing argument. Denote the Fourier coefficients of
f (x ) by lower case letters, and those of f ′ (x ) by upper
2. Let f (x ) = |x | for −1 ≤ x ≤ 1. case. Show that
(a) Write the Fourier series for f on [−1, 1]. nπ nπ
A 0 = 0, A n = b , and Bn = − an .
L n L
(b) Show that this series can be differentiated term by
term to yield the Fourier expansion of f ′ (x ) on Show that
[−π, π ]. 2 1
0 ≤ A2n − | An | +
(c) Determine f ′ (x ) and expand this function in a n n2
Fourier series on [−π, π ]. Compare this result with for n = 1, 2, · · · , with a similar inequality for Bn . Add
that of (b). these two inequalities to obtain
3. Let f (x ) = x sin(x ) for −π ≤ x ≤ π . 1 1 1
(| A | + | Bn |) ≤ ( A2n + Bn2 ) + 2 .
(a) Write the Fourier series for f on [−π, π ] n n 2 n
(b) Show that this series can be differentiated term Hence show that
by term and use this fact to obtain the Fourier L L
expansion of sin(x ) + x cos(x ) on [−π, π ]. |an | + |bn | ≤ ( A2 + Bn2 ) + .
2π n π(n 2 )
(c) Write the Fourier series for sin(x ) + x cos(x ) on
[−π, π ] and compare this result with that of (b). Thus show by comparison that
4. Let f (x ) = x 2 for −3 ≤ x ≤ 3.
∞
(|an | + |bn |)
X
(a) Write the Fourier series for f on [−3, 3]. n =1
(b) Show that this series can be differentiated term by converges. Finally, show that
term and use this to obtain the Fourier expansion
of 2x on [−3, 3]. |an cos(n π x / L ) + bn sin(n π x / L )| ≤ |an | + |bn |
(c) Expand 2x in a Fourier series on [−3, 3] and and apply a theorem of Weierstrass on uniform conver-
compare this result with that of (b). gence.
13.5 Phase Angle Form

A function f has period p if f (x + p) = f (x ) for all x . The smallest positive p for which
this holds is the fundamental period of f . For example sin(x ) has fundamental period 2π .
The graph of a function with fundamental period p simply repeats itself over intervals of
length p. We can draw the graph for − p/2 ≤ x < p/2, then replicate this graph on p/2 ≤ x <
3 p/2, 3 p/2 ≤ x < 5 p/2, −3 p/2 ≤ x < − p/2, and so on.
13.5 Phase Angle Form 453
Now suppose f has fundamental period p. Its Fourier series on [− p/2, p/2], with L = p/2,
is
1 ∞
a0 + (an cos(2n π x / p) + bn sin(2n π x / p)),

X
2 n =1
where
2 p /2
Z
an = f (x ) cos(2n π x / p) dx
p − p /2
and
2 p /2
Z
bn = f (x ) sin(2n π x / p) dx .
p − p /2
It is sometimes convenient to write this series in a different way. Let

2π
ω0 = .
p
The Fourier series of f (x ) on [− p/2, p/2] is
1 ∞
a0 + (an cos(n ω0 x ) + bn sin(n ω0 x )),

X
2 n =1
where
2 p/2
Z
an = f (x ) cos(n ω0 x ) dx
p − p /2
and
2 p/2
Z
bn = f (x ) sin(n ω0 x ) dx .
p − p/2
Now look for numbers cn and δn so that

an cos(n ω0 x ) + bn sin(n ω0 x ) = cn cos(n ω0 x + δn ).
To solve for these constants, use a trigonometric identity to write this equation as
an cos(n ω0 x ) + bn sin(n ω0 x ) = cn cos(n ω0 x ) cos(δn ) − cn sin(n ω0 x ) sin(δn ).
One way to satisfy this equation is to put
cn cos(δn ) = an and cn sin(δn ) = −bn .
If we square both sides of these equations and add the results, we obtain
cn2 = an2 + bn2 ,
so
cn = an2 + bn2 .
p
Next, divide to obtain

cn sin(δn ) bn
= tan(δn ) = − ,
cn cos(δn ) an
assuming that an 6= 0. Then
bn

δn = − arctan .
an
When each an 6= 0, these equations enable us to write the phase angle form of the Fourier
series of f (x ) on [− p/2, p/2]:
1 ∞
a + c cos(n ω0 x + δn ),
X
2 0 n=1 n
where
an2 + bn2 , and δn = − arctan(bn /an ).
ω0 = 2π/ p , cn =
p
This phase angle form is also called the harmonic form of the Fourier series for f (x )
on [− p/2, p/2]. The term cos(n ω0 x + δn ) is called the n th harmonic of f , cn is the n th
harmonic amplitude, and δn is the n th phase angle of f .
If f has fundamental period p, then in the expressions for the coefficients an and bn , we can
compute the integrals over any interval [α, α + p], since any interval of length p carries all of the
information about a p-periodic function.
This means that the Fourier coefficients of p-periodic f can be obtained as
2 α+ p
Z
an = f (x ) cos(n ω0 x ) dx
p α
and
2 α+ p
Z
bn = f (x ) sin(n ω0 x ) dx
p α
for any number α .
EXAMPLE 13.16
Let
f (x ) = x 2 for 0 ≤ x < 3
and suppose f has fundamental period p = 3. A graph of f is shown in Figure 13.17.
Since f is 3-periodic, and we are given an algebraic expression for f (x ) only on [0, 3), we
will use this interval to compute the Fourier coefficients of f . That is, use p = 3 and α = 0 in the
preceding discussion. We also have ωo = 2π/ p = 2π/3.
The Fourier coefficients are
2 3
Z
a0 = x 2 dx = 6,
3 0
2 3
2n π x 9
Z
an = x cos
2
dx = 2 2 ,
3 0 3 nπ
13.5 Phase Angle Form 455
0
–2 0 2 4 6
x
FIGURE 13.17 f (x ) in Example 13.16.
and
2 3
2n π x 9
Z
bn = x sin
2
dx = − .
3 0 3 nπ
The Fourier series of f (x ) is
∞
9 1 2n π x 2n π x

3+ cos − sin
X
.
n =1
nπ nπ 3 3
We may think of this as the Fourier series of f (x ) on the symmetric interval [−3/2, 3/2]. How-
ever, keep in mind that f (x ) is not x 2 on this interval. We have f (x ) = x 2 on 0 ≤ x < 3, hence
also on [0, 3/2]. But from Figure 13.17, f (x ) = (x + 3)2 on [−3/2, 0).
This Fourier series converges to
9/ 4 for x = ±3/2,


9/2 for x = 0,


(x + 3)2
 for −3/2 < x < 0,
 2
x for 0 < x < 3/2.

For the phase angle form, compute

9 √
cn = an2 + bn2 = 1 + n2π 2
p
n2π 2
and
−9/ n π

δn = arctan − = arctan(n π ) = 0.
9/ n 2 π 2
The phase angle form of the Fourier series of f (x ) is
∞
9 √ 2n π x

3+ 1 + n 2 π 2 cos
X
.
n =1
n2π 2 3
cn/2
1.5
.725
.36
ω0 2ω 0 3ω 0 4ω 0 nω 0
FIGURE 13.18 Amplitude spectrum of f

in Example 13.16.
The amplitude spectrum of a periodic function f is a plot of points (n ω0 , cn /2) for n =

1, 2, · · · , and also the point (0, |c0 |/2). For the function of Example 13.14, this is a plot of
points (0, 3) and, for nonzero integer n , points
2n π 9 √

, 1+n π .
2 2
3 2n 2 π 2
The amplitude spectrum for the function of Example 13.16 is shown in Figure 13.18, with
the intervals on the horizontal axis of length ω0 = 2π/3. This graph displays the relative effects
of the harmonics in the function. This is useful in signal analysis.

In Problems 1, 2, and 3, let f be periodic of period p. 5. Let
1 for 0 ≤ x < 1,
(
1. If g is also periodic of period p, show that α f + β g is f (x ) =
0 for 1 < x < 2,
periodic of period p, for any numbers α and β .
2. Let α be a positive number. Show that g(t ) = f (α t ) has and let f has fundamental period 2.
period p/α and h (t ) = f (t /α) has period α p. 6. Let f (x ) = 3x 2 for 0 ≤ x < 4 and let f have funda-
mental period 4.
3. If f is differentiable, show that f ′ has period p.
7. Let
In each of Problems 4 through 12, find the phase angle f (x ) = 1 + x for 0 ≤ x < 3,
(
form of the Fourier series of the function and plot some 2 for 2 ≤ x < 4,
points of the amplitude spectrum. Some of these functions
are specified by a graph. and suppose f has fundamental period 4.
8. f (x ) = cos(π x ) for 0 ≤ x < 1 and f has fundamental
4. Let f (x ) = x for 0 ≤ x < 2, with fundamental period 2. period 1.
13.6 Complex Fourier Series 457
9. f has the graph of Figure 13.19. 11. f has the graph of Figure 13.21.
y y
1
2
x
–3 –2 –1 1 2 3
1
–1
x
–3 –1 1 3
FIGURE 13.19 f (x ) in Problem 9,

Section 13.5.
FIGURE 13.21 f (x ) in Problem 11,
Section 13.5.
12. f has the graph of Figure 13.22.

10. f has the graph of Figure 13.20. y
y
k
k
x x
–2 2 4 –2 –1 0 1 2 3
FIGURE 13.20 f (x ) in Problem 10, FIGURE 13.22 f (x ) in Problem 12,

Section 13.5. Section 13.5.
13.6 Complex Fourier Series

There is a complex form of Fourier series that is sometimes used. As preparation for this, recall
that, in polar coordinates, a complex number (point in the plane) can be written
z = x + iy = r cos(θ ) + ir sin(θ )
where
r = |z | = x 2 + y 2
p
and θ is an argument of z . This is the angle (in radians) between the positive x − axis and the
line from the origin through (x , y ), or this angle plus any integer multiple of 2π . Using Euler’s
formula, we obtain the polar form of z :
z = r [cos(θ ) + i sin(θ )] = rei θ .
Now
ei θ = cos(θ ) + i sin(θ ),
and by replacing θ with −θ , we get

e−i θ = cos(θ ) − i sin(θ ).
Solve these equations for cos(θ ) and sin(θ ) to obtain the complex exponential forms of the
trigonometric functions:
1 1
cos(θ ) = ei θ + e−i θ , sin(θ ) = ei θ − e−i θ .

2 2i
We will also use the fact that, if x is a real number, then the conjugate of ei x is
ei x = e−i x .
This follows from Euler’s formula, since
ei x = cos(x ) + i sin(x ) = cos(x ) − i sin(x ) = e−i x .
Now let f be a piecewise smooth periodic function with fundamental period 2 L . To derive a
complex Fourier expansion of f (x ) on [− L , L ], begin with the Fourier series of f (x ). With
ω0 = π/ L , this series is
1 ∞
a0 + (an cos(n ω0 x ) + bn sin(n ω0 x ))

X
2 n =1
Put the complex forms of cos(n ω0 x ) and sin(n ω0 x ) into this expansion:
1 X ∞
1 inω0 x −inω0 x 1 inω0 x −inω0 x

a+ a e +e + bn e −e
2 0 n=1 n 2 2i
1 ∞
1 1

= a0 + (an − ibn )einω0 x + (an + ibn )e−inω0 x
X
,
2 n =1
2 2
in which we used the fact that 1/ i = −i . In this series, let
1
d0 = a0
2
and, for n = 1, 2, · · · ,
1
dn = (an − ibn ).
2
The Fourier series on [− L , L ] becomes
∞ ∞
d0 + dn e in ω0 x
dn e−inω0 x (13.14)
X X
+
n =1 n =1
Now
1 1 L
Z
d0 = a 0 = f (x ) dx
2 L −L
and for n = 1, 2, · · · ,
1
dn = (an − ibn )
2
1 L i L
Z Z
= f (x ) cos(n ω0 x ) dx − f (x ) sin(n ω0 x ) dx
2 L −L 2 L −L
1 L
Z
= f (x )[cos(n ω0 x ) − i sin(n ω0 x )] dx
2 L −L
1 L
Z
= f (x )e−inω0 x dx .
2 L −L
13.6 Complex Fourier Series 459
Then
1 L 1 L
Z Z
dn =
f (x )e−inω0 x dx = f (x )einω0 x dx = d−n .
2 L −L 2 L −L
With this, the expansion of equation (13.14) becomes
∞ ∞
d0 + dn einω0 x + d−n e−inω0 x

X X
n =1 n =1
∞
dn einω0 x .
X
=
n =−∞
This leads us to define the complex Fourier series of f on [− L , L ] to be

∞
dn einω0 x ,
X
n =−∞
with coefficients
1 L
Z
dn = f (x )e−inω0 x dx
2L −L
for n = 0, ±1, ±2, · · · .
Because of the periodicity of f , the integral defining the coefficients can be carried out over
any interval [α, α + 2 L ] of length 2 L . The Fourier convergence theorem applies to this complex
Fourier expansion, since it is just the Fourier series in complex form.
EXAMPLE 13.17
Let f (x ) = x for −1 ≤ x < 1 and suppose f has fundamental period 2, so f (x + 2) = f (x ) for

all x . Figure 13.23 is part of a graph of f . Now ω0 = π .
Immediately d0 = 0 because f is an odd function. For n 6= 0,
1 1 −inπ x
Z
dn = xe dx
2 −1
1
= 2 2 in π einπ − einπ + in π e−inπ + e−inπ

2n π
1
= 2 2 in π einπ + e−inπ − einπ − e−inπ .

2n π
The complex Fourier series of f is
∞
1
in π einπ + e−inπ − einπ − e−inπ einπ x .
X
n =−∞,n 6=0
2n π
2 2
This converges to x for −1 < x < 1. In this example we can simplify the series. For n 6= 0,
1
dn = 2 2 [2in π cos(n π ) − 2i sin(n π )]
2n π
i
= (−1)n .
nπ
0.5
0
–3 –2 –1 0 x 1 2 3
–0.5
–1
FIGURE 13.23 Graph of f in Example 13.17.
1
All the terms sin(n π ) = 0. In −n=−∞ , replace n with −n and sum from n = 1 to ∞, then combine
P
the two summations from 1 to ∞ to write
∞
i
(−1)n einπ x
X
n =−∞,n 6=0
n π
∞
i i

(−1)n einπ x + (−1)−n e−inπ x
X
=
n =1
nπ −n π
∞
i
(−1)n einπ x − e−inπ x
X
=
n =1
nπ
∞
2
(−1)n+1 sin(n π x ).
X
=
n =1
nπ
This is the Fourier series for f (x ) = x on [−1, 1].
The amplitude spectrum of a complex Fourier series of a periodic function is a graph of the
points (n ω0 , |dn |). Sometimes this graph is also referred to as a frequency spectrum.
In each of Problems 1 through 7, write the complex Fourier 4. f (x ) = 1 − x for 0 ≤ x < 6, period 6
series of f , determine the sum of the series, and plot some (
−1 for 0 ≤ x < 2
points of the frequency spectrum. 5. f (x ) = , f has period 4
2 for 2 ≤ x < 4
1. f (x ) = 2x for 0 ≤ x < 3, period 3
6. f (x ) = e−x for 0 ≤ x < 5, period 5
2. f (x ) = x 2 for 0 ≤ x < 2, period 2
x for 0 ≤ x < 1
(
0 for 0 ≤ x < 1 7. f (x ) = , f has period 2

(
3. f (x ) = , f has period 4 2 − x for 1 ≤ x < 2
1 for 1 ≤ x < 4
13.7 Filtering of Signals 461
13.7 Filtering of Signals
A periodic signal f (t ) of period 2 L is sometimes filtered to cancel out or diminish unwanted

effects, or perhaps to enhance other effects. We will briefly examine one way this is done.
Suppose f has complex Fourier series
∞
dn enπ it /L ,
X
n =−∞
where
1 L
Z
dn = f (t )e−nπ it /L dt .
2L −L
The N th partial sum of the series is

N
S N (t ) = d j eπ i jt /L .
X
j =− N
A filtered partial sum of this Fourier series is a sum of the form

N
j

Z d j eπ i jt /L . (13.15)
X
j =− N
N
Z is the filter function and is assumed to be a continuous even function on [− L , L ]. In
applications the object is to choose Z to achieve some specified purpose or effect.
To illustrate, we will develop a filter that damps out the Gibbs phenomenon. In the nineteenth
century there was an intense effort to understand convergence properties of Fourier series. In the
course of this work it was observed that the sequence of averages of partial sums of a Fourier
series is in general better behaved than the sequence of partial sums of the series itself. If SN is
the N th partial sum of the series, this average has the form
N −1
1
σ N (t ) = Sk (t ).
X
N k =0
The quantity σRN (t ) is called the N th Cesàro sum of f . It was shown that, if f is periodic of
period 2π and 02π f (t ) dt exists, then σ N (t ) → f (t ) for any t at which f is continuous, a much
stronger result than holds for partial sums of Fourier series.
Inserting the summation for Sk (t ), we have
N −1 X
k
1
!
σ N (t ) = d j eπ i jt /L .
X
N k =0 j =−k
With some manipulation, this double sum can be rearranged to write

N n
σ N (t ) = 1 − dn eπ int / L .
X
n =− N
N
This is of the form of equation (13.15) with the Cesàro filter function
Z (t ) = 1 − |t | for − 1 ≤ t ≤ 1.
The sequence
h n i N n i N
1 −
h
Z =

N N

n =− N n =− N
is called the sequence of filter factors for the Cesàro filter.
This “averaging” filter damps out the Gibbs effect in the convergence of a Fourier series. To
observe this, let f have fundamental period 2π , and, on [−π, π],
−1 for −π ≤ t < 0
(
f (t ) =
1 for 0 ≤ t < π .
The complex Fourier series of f is
∞
i −1 + (−1)n int
e .
X
n =−∞,n 6=0
π n
The N th partial sum of this series is
N
i −1 + (−1)n int
S N (t ) = e .
X
n =− N ,n 6=0
π n
If we pair positive and negative values of n in this summation, we find that
N
−2
SN (t ) = (−1 + (−1)n ) sin(nt ).
X
n =1
nπ
If N is even, then −1 + (−1)n = 0. If N is odd, then −1 + (−1)n = −2. Therefore, for odd N ,
4 1 1 1

S N (t ) = sin(t ) + sin(3t ) + sin(5t ) + · · · + sin( Nt ) .
π 3 5 N
The N th Cesàro sum is
N n i −1 + (−1)n
1 −

σ N (t ) = eint .
X
n =− N ,n 6=0
N π n
With some manipulation, this can be written

N
n −2 −1 + (−1)n

σ N (t ) = 1− sin(nt ).
X
n =1
N π n
Figure 13.24 shows graphs of S10 (t ) and σ10 (t ), and Figure 13.25 graphs of S30 (t ) and σ30 (t ),
showing the Gibbs effect in the partial sums of the Fourier series, and this effect damped out in
the smoother Cesàro sums. The Cesàro filter also damps out the higher frequency terms in the
Fourier series because 1 − |n / N | tends to zero as n increases toward N .
13.7 Filtering of Signals 463
1 1
0.5 0.5
0 –3 –2 –1 0 1 2 3
–3 –2 –1 0 1 2 3 0
–0.5 –0.5
–1 –1
FIGURE 13.24 Tenth partial sum and Cesàro FIGURE 13.25 Thirtieth partial sum and
sum of f . Cesàro sum of f .
There are many filters used in signal analysis. Two of the more frequently used ones are the
Hamming and Gauss filters.
The Hamming filter is named for Richard Hamming, who was a senior research scientist
at Bell Labs, and is defined by
Z (t ) = 0.54 + 0.46 cos(π t ).
The Gauss filter is sometimes used to filter out background noise and is defined by
22
Z (t ) = e−απ t ,
with α a positive constant.

In each of Problems 1 through 5, graph the function, the −1 for −1 ≤ t < −1/2


fifth partial sum of its Fourier series on the interval, and 3. f (t ) = 0 for −1/2 ≤ t < 1/2
the fifth Cesàro sum, using the same set of axes. Repeat this 

1 for 1/2 ≤ t < 1
process for the tenth and twenty-fifth partial sums. Notice
in particular the graphs at points of discontinuity of the
function, where the Gibbs phenomenon appears. 0 for −3 ≤ t < 0
(
4. f (t ) =
cos(t ) for 0 ≤ t < 3
1 for 0 ≤ t < 2
(
1. f (t ) =
−1 for −2 ≤ t < 0 2 + t for −1 ≤ t < 0
(
5. f (t ) =
t2 for −2 ≤ t < 1 7 for 0 ≤ t < 1
(
2. f (t ) =
2 + t for 1 ≤ t < 2
6. Let 7. Let
for 0 ≤ t < 2
1 t for −2 ≤ t < 0
( (
f (t ) = f (t ) =
−1 for −2 ≤ t < 0 2 + t for 0 ≤ t < 2
Plot the fifth partial sum of the Fourier series for f (t ) Plot the fifth partial sum of the Fourier series for f (t )
on [−2, 2], together with the fifth Cesàro sum, the on [−2, 2], together with the fifth Cesàro sum, the
fifth Hamming and Gauss filtered partial sums, using fifth Hamming and Gauss filtered partial sums, using
the same set of axes. Repeat this with the tenth and the same set of axes. Repeat this with the tenth and
twenty-fifth partial sums. twenty-fifth partial sums.

13.1 Why Fourier Series?

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

13.1 Why Fourier Series?

Uploaded by

Copyright:

Available Formats

WHY FOURIER SERIES?

13.1 Why Fourier Series?

A Fourier series is a representation of a function as a series of constant multiples of sine and/or

Fourier found that the functions

But this would require that N and numbers b1 , · · · , b N be found so that

SECTION 13.1 PROBLEMS

1. Let 2. Let p(x ) be a polynomial. Prove that there is no number

13.2 The Fourier Series of a Function

Let f (x ) be defined on [− L , L ]. We want to choose numbers a0 , a1 , a2 · · · and b1 , b2 , · · ·

f (x ) = a0 + [ak cos(k π x / L ) + bk sin(k π x / L )]. (13.1)

by equation (13.5). Therefore

Let f (x ) = x − x 2 for −π ≤ x ≤ π . Here L = π . Compute

4 sin(n π ) − 4n π cos(n π ) − 2n 2 π 2 sin(n π )

FIGURE 13.1 A piecewise con-

FIGURE 13.2 f in Example 13.2.

f is piecewise smooth on [a , b] if f is piecewise continuous and f ′ exists and is continuous

The function f of Example 13.2 is differentiable on (−3, 4) except at x = 2:

We can now state a convergence theorem.

THEOREM 13.1 C o nve r g e n c e o f Fo u r i e r S e r i e s

Let f be piecewise smooth on [− L , L ]. Then, for each x in (− L , L ), the Fourier series of f on

FIGURE 13.3 Convergence of a Fourier

At both − L and L this Fourier series converges to

Let f (x ) = ex . The Fourier coefficients of f on [−1, 1] are

FIGURE 13.6 Twentieth partial sum in

FIGURE 13.7 Thirtieth partial sum of the

5 sin(x ) for −2π ≤ x < −π/2

13.2.1 Even and Odd Functions

A function f is even on [− L , L ] if its graph on [− L , 0] is the reflection across the vertical

and if f is odd on [− L , L ] then

FIGURE 13.10 Twentieth partial sum of the

FIGURE 13.11 Tenth partial sum of the

FIGURE 13.12 The Gibbs phenomenon.

SECTION 13.2 PROBLEMS

−x for −5 ≤ x < 0 2x − 2 for −π ≤ x ≤ 1

1 for −π ≤ x < 0 x 2 for −π ≤ x ≤ 0

cos(x ) for −2 ≤ x < 0 −2 for −4 ≤ x ≤ −2

13.3 Sine and Cosine Series

FIGURE 13.13 Even extension of a

in which the an s are the Fourier cosine coefficients of f on [0, L ].

THEOREM 13.2 C o nve r g e n c e o f Fo u r i e r C o s i n e S e r i e s

Let f be piecewise smooth on [0, L ]. Then

The cosine series for e2x on [0, 1] is

This series converges to

FIGURE 13.14 Fifth partial sum of the

FIGURE 13.15 Odd extension of a func-

The Fourier sine coefficients of f on [0, L ] are

is the Fourier sine series for f on [0, L ].

THEOREM 13.3 C o nve r g e n c e o f Fo u r i e r S i n e S e r i e s

Let f be piecewise smooth on [0, L ]. Then

FIGURE 13.16 Fortieth partial sum of the sine

SECTION 13.3 PROBLEMS

13.4 Integration and Differentiation of Fourier Series

for −π < x < π . Differentiate this series term by term to get

This series does not converge on (−π, π ). 

THEOREM 13.4 Integration of Fourier Series

Let f be piecewise continuous on [− L , L ], with Fourier series

a0 + (an cos(n π x / L ) + bn sin(n π x / L )).

Then, for any x with − L ≤ x ≤ L ,

From Example 13.11,

THEOREM 13.5 Differentiation of Fourier Series

Let f be continuous on [− L , L ] and suppose that f (− L ) = f ( L ). Let f ′ be piecewise

f (x ) = a0 + (an cos(n π x / L ) + bn sin(n π x / L ))

for − L ≤ x ≤ L . Further, at each x in (− L , L ) at which f ′′ (x ) exists, the term by term derivative

This series does not converge on (−π, π ).

This is the Fourier series for f (x ) = x on [−1, 1].