Fourier Series PDF

Chapter 14
Fourier Series
14.1 General Properties
Periodic phenomena involving waves [∼ sin(2π x/λ) as a crude approximation

to water waves, for example], motors, rotating machines (harmonic motion),
or some repetitive pattern of a driving force are described by periodic func-
tions. Fourier series are a basic tool for solving ordinary differential equa-
tions (ODEs) and partial differential equations (PDEs) with periodic bound-
ary conditions. Fourier integrals for nonperiodic phenomena are developed in
Chapter 15. The common name for the whole field is Fourier analysis.
A Fourier series is defined as an expansion of a function or representation
of a function in a series of sines and cosines, such as
a0 ∞ ∞
f (x) = + a n cos nx + bn sin nx. (14.1)
2 n=1 n=1
The coefficients a 0 , a n, and bn are related to the periodic function f (x) by

definite integrals:
2π
1
an = f (x) cos nx dx, (14.2)
π 0
2π
1
bn = f (x) sin nx dx, n = 0, 1, 2, . . . . (14.3)
π 0
This result, of course, is subject to the requirement that the integrals exist.
They do if f (x) is piecewise continuous (or square integrable). Notice that a 0
is singled out for special treatment by the inclusion of the factor 12 . This is done
so that Eq. (14.2) will apply to all a n, n = 0 as well as n > 0.
The Sturm–Liouville theory of the harmonic oscillator in Example 9.1.4
guarantees the validity of Eqs. (14.2) and (14.3) and, by use of the orthogonality
relations (Example 9.2.1), allows us to compute the expansion coefficients.
663
664 Chapter 14 Fourier Series
Another way of describing what we are doing here is to say that f (x) is part
of an infinite-dimensional Hilbert space, with the orthogonal cos nx and sin nx
as the basis. The statement that cos nx and sin nx (n = 0, 1, 2, . . .) span this
Hilbert space is equivalent to saying that they form a complete set. Finally, the
expansion coefficients a n and bn correspond to the projections of f (x),
with the integral inner products [Eqs. (14.2) and (14.3)] playing the role of the
dot product of Section 1.2. These points are outlined in Section 9.4.
The conditions imposed on f (x) to make Eq. (14.1) valid, and the series
convergent, are that f (x) has only a finite number of finite discontinuities and
only a finite number of extreme values, maxima, and minima in the interval
[0, 2π].1 Functions satisfying these conditions are called piecewise regular.
The conditions are known as the Dirichlet conditions. Although there are
some functions that do not obey these Dirichlet conditions, they may well be
labeled pathological for purposes of Fourier expansions. In the vast majority of
physical problems involving a Fourier series, these conditions will be satisfied.
In most physical problems we shall be interested in functions that are square
integrable, for which the sines and cosines form a complete orthogonal set.
This in turn means that Eq. (14.1) is valid in the sense of convergence in the
mean (see Eq. 9.63).
Completeness
The Fourier expansion and the completeness property may be expected be-
cause the functions sin nx, cos nx, ein x are all eigenfunctions of a self-adjoint
linear ODE,
y + n2 y = 0. (14.4)
We obtain orthogonal eigenfunctions for different values of the eigenvalue n
for the interval [0, 2π ] to satisfy the boundary conditions in the Sturm–Liouville
theory (Chapter 9). The different eigenfunctions for the same eigenvalue n are
orthogonal. We have
2π
π δmn, m = 0,
sin mx sin nx dx = (14.5)
0 0, m = 0,

2π π δmn, m= 0,
cos mx cos nx dx = (14.6)
0 2π, m = n = 0,
2π
sin mx cos nx dx = 0, for all integral m and n. (14.7)
0
Note that any interval x0 ≤ x ≤ x0 + 2π will be equally satisfactory. Fre-

quently, we use x0 = −π to obtain the interval −π ≤ x ≤ π . For the complex
eigenfunctions e±in x , orthogonality is usually defined in terms of the complex
1 These conditions are sufficient but not necessary.

14.1 General Properties 665
conjugate of one of the two factors,

2π
(eimx )∗ ein x dx = 2π δmn. (14.8)
0
This agrees with the treatment of the spherical harmonics (Section 11.5).
EXAMPLE 14.1.1 Sawtooth Wave Let us apply Eqs. (14.2) and (14.3) to the sawtooth shape
shown in Fig. 14.1 to derive its Fourier series. Our sawtooth function can also
be expressed as

x, 0 ≤ x < π,
f (x) =
x − 2π, π ≤ x ≤ 2π,
which is an odd function of the variable x. Hence, we expect a pure sine

expansion. Integrating by parts, we indeed find
π π
1 π x 1
an = x cos nx dx =
sin nx − sin nx dx
π −π nπ −π nπ −π
π
1 1
= 2 cos nx = 2 [(−1)n − (−1)n] = 0,
nπ −π n π
while
π π
1 π
x 1
bn = x sin nx dx = −
cos nx + sin nx dx
π −π nπ −π nπ −π
π
2 1 2
= − (−1)n − 2 cos nx = − (−1)n.
n nπ −π n
This establishes the Fourier expansion

sin 2x sin 3x n+1 sin nx
f (x) = 2 sin x − + − · · · + (−1) + ···
2 3 n
∞
sin nx
=2 (−1)n+1 = x, − π < x < π, (14.9)
n=1
n
Figure 14.1
y
Sawtooth Wave
p
0 x
–2p –p p 2p
–p
which converges only conditionally, not absolutely, because of the disconti-

nuity of f (x) at x = ±π. It makes no difference whether a discontinuity is
in the interior of the expansion interval or at its ends: It will give rise to con-
ditional convergence of the Fourier series. In terms of physical applications
with x = ω a frequency, conditional convergence means that our square wave
is dominated by high-frequency components. ■
Behavior of Discontinuities
The behavior at x = nπ is an example of a general rule that at a finite dis-
continuity the series converges to the arithmetic mean. For a discontinuity at
x = x0 the series yields
1
f (x0 ) = [ f (x0 + 0) + f (x0 − 0)], (14.10)
2
where the arithmetic mean of the right and left approaches to x = x0 . A general
proof using partial sums is given by Jeffreys and by Carslaw (see Additional
Reading).
An idea of the convergence of a Fourier series and the error in using only
a finite number of terms in the series may be obtained by considering the
expansion of the sawtooth shape of Fig. 14.1. Figure 14.2 shows f (x) for 0 ≤
x < π for the sum of 4, 6, and 10 terms of the series. Three features deserve
comment:
1. There is a steady increase in the accuracy of the representation as the
number of terms included is increased.
2. All the curves pass through the midpoint f (x) = 0 at x = π .
3. In the vicinity of x = π there is an overshoot that persists and shows no
sign of diminishing. This overshoot (and undershoot) is called the Gibbs
phenomenon and is a typical feature of Fourier series. The inclusion of
more terms does nothing to remove the overshoot (undershoot) but merely
moves it closer to the point of discontinuity. The Gibbs phenomenon is not
Figure 14.2
f(x)
Fourier 6
10 terms
Representation of 4
Sawtooth Wave
x
p
Figure 14.3
f(t)
Full Wave Rectifier
wt
–2p –p p 2p
limited to Fourier series. It occurs with other eigenfunction expansions. For

more details, see W. J. Thomson, Fourier series and the Gibbs phenomenon,
Am. J. Phys. 60, 425 (1992).
One of the advantages of a Fourier representation over some other

representation, such as a Taylor series, is that it can represent a dis-
continuous function. An example is the sawtooth wave in the preceding
section and Example 14.1.3. Other examples are considered in the exercises.
EXAMPLE 14.1.2 Full-Wave Rectifier Consider the case of an absolutely convergent Fourier
series representing a continuous periodic function, displayed in Fig. 14.3. Let
us ask how well the output of a full-wave rectifier approaches pure direct
current. Our rectifier may be thought of as having passed the positive peaks
of an incoming sine wave and inverting the negative peaks. This yields
f (t) = sin ωt, 0 < ωt < π,

(14.11)
f (t) = − sin ωt, −π < ωt < 0.
Since f (t) defined here is even, no terms of the form sin nωt will appear. Again,
from Eqs. (14.2) and (14.3), we have

1 0 1 π
a0 = − sin ωt d(ωt) + sin ωt d(ωt)
π −π π 0

2 π 4
= sin ωt d(ωt) = , (14.12)
π 0 π

2 π
an = sin ωt cos nωt d(ωt)
π 0
2 2
=− , n even,
π n2 − 1
= 0, n odd. (14.13)
Figure 14.4
f(x)
Square Wave
x
–3p –2p –p p 2p 3p
Note that (0, π ) is not an orthogonality interval for both sines and cosines
together and we do not get zero for even n. The resulting series is
2 4
∞
cos nωt
f (t) = − . (14.14)
π π n=2,4,6,...
n2 − 1
The original frequency ω has been eliminated. The lowest frequency oscillation
is 2ω. The high-frequency components fall off as n−2 , showing that the full-wave
rectifier does a fairly good job of approximating direct current. Whether this
good approximation is adequate depends on the particular application. If the
remaining ac components are objectionable, they may be further suppressed
by appropriate filter circuits. These two examples highlight two features char-
acteristic of Fourier expansions:2
• If f (x) has discontinuities (as in the sawtooth wave in Example 14.1.1), we
can expect the nth coefficient to be decreasing as O(1/n). Convergence is
conditional only.3
• If f (x) is continuous (although possibly with discontinuous derivatives as in
the full-wave rectifier of Example 14.1.2), we can expect the nth coefficient
to be decreasing as 1/n2 , that is, absolute convergence. ■
EXAMPLE 14.1.3 Square Wave–High Frequencies One application of Fourier series, the
analysis of a “square” wave (Fig. 14.4) in terms of its Fourier components,
occurs in electronic circuits designed to handle sharply rising pulses. This
example explains the physical meaning of conditional convergence. Suppose
that our wave is defined by
f (x) = 0, −π < x < 0,

(14.15)
f (x) = h, 0 < x < π.
2 Raisbeek, G. (1955). Order of magnitude of Fourier coefficients. Am. Math. Mon. 62, 149–155.
3A technique for improving the rate of convergence is developed in the exercises of Section 5.9.
From Eqs. (14.2) and (14.3) we find

1 π
a0 = h dt = h, (14.16)
π 0

1 π
an = h cos nt dt = 0, n = 1, 2, 3, . . . , (14.17)
π 0

1 π h
bn = h sin nt dt = (1 − cos nπ ); (14.18)
π 0 nπ
2h
bn = , n odd, (14.19)
nπ
bn = 0, n even. (14.20)
The resulting series is

h 2h sin x sin 3x sin 5x
f (x) = + + + + ··· . (14.21)
2 π 1 3 5
Except for the first term, which represents an average of f (x) over the in-
terval [−π, π ], all the cosine terms have vanished. Since f (x) − h/2 is odd,
we have a Fourier sine series. Although only the odd terms in the sine series
occur, they fall only as n−1 . This conditional convergence is like that of the
harmonic series. Physically, this means that our square wave contains a lot of
high-frequency components. If the electronic apparatus will not pass these
components, our square wave input will emerge more or less rounded off,
perhaps as an amorphous blob. ■
Biographical Data
Fourier, Jean Baptiste Joseph, Baron. Fourier, a French mathematician,
was born in 1768 in Auxerre, France, and died in Paris in 1830. After his grad-
uation from a military school in Paris, he became a professor at the school in
1795. In 1808, after his great mathematical discoveries involving the series
and integrals named after him, he was made a baron by Napoleon. Earlier,
he had survived Robespierre and the French Revolution. When Napoleon re-
turned to France in 1815 after his abdication and first exile to Elba, Fourier
rejoined him and, after Waterloo, fell out of favor for a while. In 1822, his
book on the Analytic Theory of Heat appeared and inspired Ohm to new
thoughts on the flow of electricity.
SUMMARY Fourier series are finite or infinite sums of sines and cosines that describe
periodic functions that can have discontinuities and thus represent a wider
class of functions than we have considered so far. Because sin nx, cos nx are
eigenfunctions of a self-adjoint ODE, the classical harmonic oscillator equa-
tion, the Hilbert space properties of Fourier series are consequences of the
Sturm–Liouville theory.
EXERCISES
14.1.1 A function f (x) (quadratically integrable) is to be represented by a

finite Fourier series. A convenient measure of the accuracy of the
series is given by the integrated square of the deviation
2π
2
a0
p

p = f (x) − − (a n cos nx + bn sin nx) dx.
0 2 n=1
Show that the requirement that

p be minimized, that is,
∂
p ∂
p
= 0, = 0,
∂a n ∂bn
for all n, leads to choosing a n and bn, as given in Eqs. (14.2) and (14.3).
Note. Your coefficients a n and bn are independent of p. This indepen-
dence is a consequence of orthogonality and would not hold for powers
of x, fitting a curve with polynomials.
14.1.2 In the analysis of a complex waveform (ocean tides, earthquakes, mu-
sical tones, etc.) it might be more convenient to have the Fourier series
written as
a0 ∞
f (x) = + α n cos(nx − θ n).
2 n=1
Show that this is equivalent to Eq. (14.1) with
a n = α n cos θ , α 2n = a2n + bn2 ,

bn = α n sin θ , tan θ n = bn/a n.
Note. The coefficients αn2 as a function of n define what is called the

power spectrum. The importance of αn2 lies in its invariance under a
shift in the phase θ n.
π
14.1.3 Assuming that −π [ f (x)]2 dx is finite, show that
lim a m = 0, lim bm = 0.
m→∞ m→∞
Hint. Integrate [ f (x) − s n(x)]2 , where s n(x) is the nth partial sum, and
use Bessel’s inequality (Section 9.4). For our finite interval the assump-
π
π that f (x) is square integrable ( −π | f (x)| dx is finite) implies that
2
tion
−π | f (x)| dx is also finite. The converse does not hold.
14.1.4 Apply the summation technique of this section to show that

1
∞
sin nx 2
(π − x), 0<x≤π
=
n=1
n − 1 (π + x), −π ≤ x < 0
2
(Fig. 14.5).
14.2 Advantages and Uses of Fourier Series 671
Figure 14.5
f (x)
Reverse Sawtooth
Wave
p
f (x) = Σ 1n sin nx
n=1
2
x
–p p
p
2
14.1.5 Sum the trigonometric series

∞
sin(2n + 1)x
,
n=0
2n + 1
and show that it equals

π/4, 0<x<π
−π/4, −π < x < 0.
14.2 Advantages and Uses of Fourier Series
Periodic Functions
Related to the advantage of describing discontinuous functions is the useful-
ness of a Fourier series in representing a periodic function. If f (x) has a period
of 2π , perhaps it is only natural that we expand it in a series of functions with
period 2π, 2π/2, 2π/3, . . . . This guarantees that if our periodic f (x) is repre-
sented over one interval [0, 2π ] or [−π, π], the representation holds for all
finite x.
At this point, we may conveniently consider the properties of symmetry.
Using the interval [−π, π], sin x is odd and cos x is an even function of x.
Hence, by Eqs. (14.2) and (14.3),4 if f (x) is odd, all a n = 0, and if f (x) is even,
4 With the range of integration −π ≤ x ≤ π .

all bn = 0. In other words,

a0 ∞
f (x) = + a n cos nx, f (x) even, (14.22)
2 n=1

∞
f (x) = bn sin nx, f (x) odd. (14.23)
n=1
Frequently, these properties are helpful in expanding a given function.

We have noted that the Fourier series is periodic. This is important in
considering whether Eq. (14.1) holds outside the initial interval. Suppose we
are given only that
f (x) = x, 0≤x<π (14.24)
and are asked to represent f (x) by a series expansion. Let us take three of the
infinite number of possible expansions:
1. If we assume a Taylor expansion, we have
f (x) = x, (14.25)
a one-term series. This (one-term) series is defined for all finite x.

2. Using the Fourier cosine series [Eq. (14.22)], we predict that
f (x) = −x, −π < x ≤ 0,
(14.26)
f (x) = 2π − x, π < x ≤ 2π.
3. Finally, from the Fourier sine series [Eq. (14.23)], we have
f (x) = x, −π < x ≤ 0,
(14.27)
f (x) = x − 2π, π < x ≤ 2π.
These three possibilities—Taylor series, Fourier cosine series, and Fourier
sine series—are each perfectly valid in the original interval [0, π ]. Outside,
however, their behavior is strikingly different (compare Fig. 14.6). Which
of the three, then, is correct? This question has no answer, unless we are
given more information about f (x). It may be any of the three or none of
them. Our Fourier expansions are valid over the basic interval. Unless the
function f (x) is known to be periodic, with a period equal to our basic in-
terval, or (1/n)th of our basic interval, there is no assurance whatever that
the representation [Eq. (14.1)] will have any meaning outside the basic inter-
val. Clearly, the interval of length 2π, which defines the expansion, makes a
real difference for a nonperiodic function because the Fourier series repeats
the pattern of the basic interval in adjacent intervals. This also follows from
Fig. 14.6.
In addition to the advantages of representing discontinuous and periodic
functions, there is a third very real advantage in using a Fourier series. Suppose
that we are solving the equation of motion of an oscillating particle, subject to
Figure 14.6
f (x)
Comparison of Fourier
Cosine Series, Fourier
Sine Series, and Taylor
Taylor series
Series
p
Fourier cosine series
x
–2p –p p 2p
Fourier sine series
a periodic driving force. The Fourier expansion of the driving force then gives
us the fundamental term and a series of harmonics. The (linear) ODE may be
solved for each of these harmonics individually, a process that may be much
easier than dealing with the original driving force. Then, as long as the ODE
is linear, all the solutions may be added together to obtain the final solution.5
This is more than just a clever mathematical trick.
• It corresponds to finding the response of the system to the funda-

mental frequency and to each of the harmonic frequencies, called Fourier
analysis.
The following question is sometimes raised: Were the harmonics there all along,
or were they created by our Fourier analysis? One answer compares the func-
tional resolution into harmonics with the resolution of a vector into rectangular
components. The components may have been present in the sense that they
may be isolated and observed, but the resolution is certainly not unique. Hence,
many authorities prefer to say that the harmonics were created by our choice
of expansion. Other expansions in other sets of orthogonal functions would
give different results. For further discussion, we refer to a series of notes and
letters in the American Journal of Physics.6
5 One of the nastier features of nonlinear differential equations is that this principle of superposition
is not valid.
6 Robinson, B. L. (1953). Concerning frequencies resulting from distortion. Am. J. Phys. 21,
391; Van Name, F. W., Jr. (1954). Concerning frequencies resulting from distortion. Am. J. Phys.
22, 94.
Change of Interval
So far, attention has been restricted to an interval of length 2π. This restriction
may easily be relaxed. If f (x) is periodic with a period 2L, we may write
∞
a0 nπ x nπ x
f (x) = + a n cos + bn sin , (14.28)
2 n=1
L L
with
L
1 nπ t
an = f (t) cos dt, n = 0, 1, 2, 3, . . . , (14.29)
L −L L
L
1 nπ t
bn = f (t) sin dt, n = 1, 2, 3, . . . , (14.30)
L −L L
replacing x in Eq. (14.1) with π x/L and t in Eqs. (14.2) and (14.3) with π t/L.
[For convenience, the interval in Eqs. (14.2) and (14.3) is shifted to −π ≤ t ≤
π.] The choice of the symmetric interval [−L, L] is not essential. For f (x)
periodic with a period of 2L, any interval (x0 , x0 + 2L) will do. The choice is
a matter of convenience or literally personal preference.
If our function f (−x) = f (x) is even in x, then it has a pure cosine series
because, by substituting t → −t,
0 L
nπ t nπ t
f (t) cos dt = f (t) cos dt,
−L L 0 L
0 L
nπ t nπ t
f (t) sin dt = − f (t) sin dt
−L L 0 L
so that all bn = 0 and
L
2 nπ t
an = f (t) cos dt.
L 0 L
Similarly, if f (−x) = − f (x) is odd in x, then f (x) has a pure sine series
with coefficients

2 L nπ t
bn = f (t) sin dt.
L 0 L
EXAMPLE 14.2.1 Asymmetric Square Wave Let us derive the Fourier expansion for the
square wave in Fig. 14.7:
f (x) = h, 0 < x < 1,

f (x) = 0, 1 < x < 2L
for L ≥ 1 and h > 0. The expansion interval is [−L, L].

Figure 14.7
Asymmetric Square Wave L+ 1
2
h
2L − 1
x
0 1 2L
The geometry of the square pulse implies that its Fourier coefficients are
given by Eqs. (14.29) and (14.30) as

h 1
nπ x h nπ x 1 h nπ
an = cos dx = sin = sin , n = 0, 1, . . . ,
L 0 L nπ L 0 nπ L

h 1
nπ x h nπ x 1 h nπ
bn = sin dx = − cos = 1−cos , n = 1, 2, . . . .
L 0 L nπ L 0 nπ L
The resulting Fourier series

h ∞
1 nπ nπ x nπ nπ x
f (x) = sin cos + 1 − cos sin
π n=1 n L L L L
for L = 1 agrees with the square wave series of Example 14.1.3 because
sin nπ = 0 and cos nπ = (−1)n.
Because f is symmetric about x = 2L+12
(dashed vertical line in Fig. 14.7),
we shift the origin of our coordinates from x = 0 to this new origin, calling
ξ = x− 2L+1
2
the new variable. We expect a pure cosine series with coefficients

h nπ ξ L+1/2
h nπ ξ L+1/2
An = dξ = cos
sin
L
L−1/2 L nπ L L−1/2

h nπ(2L + 1) nπ(2L − 1) 2h nπ
= sin − sin = (−1)n sin ,
nπ 2L 2L nπ 2L
and B n = 0. The Fourier series
2h
∞
(−1)n nπ nπ(x − L − 1/2)
f (x) = sin cos
π n=1 n 2L L
can be seen to be equivalent to our first result using the addition formulas for
trigonometric functions

nπ(x − L − 1/2) nπ nπ x nπ nπ x
cos = (−1)n cos cos + sin sin . ■
L 2L L 2L L
EXERCISES
14.2.1 The boundary conditions [such as ψ(0) = ψ(l) = 0] may suggest solu-
tions of the form sin(nπ x/l) and eliminate the corresponding cosines.
(a) Verify that the boundary conditions used in the Sturm–Liouville
theory are satisfied for the interval (0, l). Note that this is only half
the usual Fourier interval.
(b) Show that the set of functions ϕn(x) = sin(nπ x/l), n = 1, 2, 3, . . .
satisfies an orthogonality relation
l
l
ϕm(x)ϕn(x)dx = δmn, n > 0.
0 2
14.2.2 (a) Expand f (x) = x in the interval [0, 2L]. Sketch the series you have
found (right-hand side of Answer) over [−2L, 2L].

2L ∞
1 nπ x
ANS. x = L − sin .
π n=1 n L
(b) Expand f (x) = x as a sine series in the half interval (0, L). Sketch
the series you have found (right-hand side of Answer) over
[−2L, 2L].
4 ∞
1 (2n + 1)π x
ANS. x = sin .
π n=0 2n + 1 L
14.2.3 In some problems it is convenient to approximate sin π x over the inter-

val [0, 1] by a parabola ax(1 − x), where a is a constant. To get a feeling
for the accuracy of this approximation, expand 4x(1 − x) in a Fourier
sine series:

4x(1 − x), 0 ≤ x ≤ 1 ∞
f (x) = = bn sin nπ x.
4x(1 + x), −1 ≤ x ≤ 0 n=1
32 1
ANS. bn = · , n odd
π 3 n3
bn = 0, n even
(Fig. 14.8).
14.2.4 Take −π ≤ x ≤ π as the basic interval for the function f (x) = x and
repeat the arguments leading to Eqs. (14.25)–(14.27). Compare your
results with those in Fig. 14.6 and plot them.
14.3 Complex Fourier Series 677
Figure 14.8
Parabolic Sine Wave f (x)
–1 1
x
14.3 Complex Fourier Series
One way of summing a trigonometric Fourier series is to transform it into

exponential form and compare it with a Laurent series (see Section 6.5). If we
expand f (z) in a Laurent series [assuming f (z) is analytic],

∞
f (z) = cn z n , (14.31)
n=−∞
then on the unit circle z = eiθ = cos θ + i sin θ by Euler’s identity so that

∞
f (z) = f (eiθ ) = cneinθ . (14.32)
n=−∞
Expressing cos nx and sin nx in exponential form and setting x = θ ,

1 inx 1 inx
cos nx = (e + e−inx ), sin nx = (e − e−inx ),
2 2i
we may rewrite the general Fourier series [Eq. (14.1)] as
∞
a0 1 1
f (x) = + a n (ein x + e−in x ) + bn (ein x − e−in x )
2 n=1
2 2i

∞
= cnein x , (14.33)
n=−∞
in which
1 1 1
cn = (a n − ibn), c−n = (a n + ibn), n > 0, c0 = a0. (14.34)
2 2 2
The Laurent expansion on the unit circle [Eq. (14.32)] has the same form as the
complex Fourier series [Eq. (14.33)], which shows the equivalence between
the two expansions.
When the function f (z) is known, the complex Fourier coefficients may be
directly derived by projection from it:
2π
1
cn = e−in x f (x)dx, − ∞ < n < ∞,
2π 0
based on the orthogonality relation [Eq. (14.8)].
Abel’s Theorem
Consider a function f (z) represented by a convergent power series

∞
∞
f (z) = cn z n = cnr neinθ . (14.35)
n=0 n=0
This is our Fourier exponential series [Eq. (14.32)]. Separating real and imag-
inary parts

∞
∞
u(r, θ) = cnr n cos nθ, v(r, θ) = cnr n sin nθ, (14.36)
n=0 n=1
the Fourier cosine and sine series. Abel’s theorem asserts that if u(1, θ) and
v(1, θ) are convergent for a given θ , then
u(1, θ ) + iv(1, θ ) = lim f (reiθ ). (14.37)

r→1
An application of this theorem appears as Exercise 14.3.11, and it is used in

the next example.
EXAMPLE 14.3.1 Summation

∞ of a Complex Fourier Series Consider the series
n=1 (1/n) cos nx, x ∈ (0, 2π ). Since this series is only conditionally conver-
gent (see Example 5.3.1) and diverges at x = 0 so that Dirichlet’s conditions
are violated, we take

∞
cos nx
∞
r n cos nx
= lim (14.38)
n=1
n r→1
n=1
n
absolutely convergent for |r| < 1. Our procedure is again to try forming a
power series by transforming the trigonometric functions into exponential
form:
∞
r n cos nx 1 ∞
r nein x 1 ∞
r ne−in x
= + . (14.39)
n=1
n 2 n=1 n 2 n=1 n
Now these power series may be identified as Maclaurin expansions of
− ln(1 − z), z = reix , re−ix [Eq. (5.65)], and

∞
r n cos nx 1
= − [ln(1 − reix ) + ln(1 − re−ix )]
n=1
n 2
= − ln[(1 + r 2 ) − 2r cos x]1/2 . (14.40)
Table 14.1 Fourier Series Reference

∞
1 − 12 (π + x), −π ≤ x < 0 Exercise 14.1.4
1. sin nx =
n =1
n 1
2 (π − x), 0≤x<π Exercise 14.3.3

∞
1 1 Example 14.1.1
2. (−1)n+1 sin nx = x, −π < x < π
n =1
n 2 Exercise 14.3.2

∞
1 −π/4, −π < x < 0 Exercise 14.1.5
3. sin(2n + 1)x =
n =0
2n + 1 +π/4, 0<x<π Eq. (14.52)
∞
cos nx |x|
4. = − ln 2 sin , −π < x < π Eq. (14.38)
n =1
n 2
∞
1 x
5. (−1)n cos nx = − ln 2 cos , −π < x < π Exercise 14.3.11
n =1
n 2
∞
1 1 |x|
6. cos(2n + 1)x = ln cot , −π < x < π (Item 5 – Item 4)
n =0
2n + 1 2 2

∞
cos nx x2 π2
7. (−1)n 2
= − , −π < x < π Example 14.4.2
n =1
n 4 12
∞
cos(2n + 1)x π π
8. = − |x| , −π < x < π Exercise 14.3.4
n =0
(2n + 1)2 4 2
Letting r = 1 based on Abel’s theorem, we obtain item 4 of Table 14.1

∞
cos nx
= − ln(2 − 2 cos x)1/2
n=1
n

x
= − ln 2 sin , x ∈ (0, 2π ).7 (14.41)
2
Both sides of this expression diverge as x → 0 and 2π . ■
EXERCISES
14.3.1 Develop the Fourier series representation of

0, −π ≤ ωt ≤ 0,
f (t) =
sin ωt, 0 ≤ ωt ≤ π.
This is the output of a simple half-wave rectifier. It is also an ap-
proximation of the solar thermal effect that produces “tides” in the
atmosphere.
1 1 2 ∞
cos nωt
ANS. f (t) = + sin ωt − .
π 2 π n=2,4,6,... n2 − 1
7 The limits may be shifted to (−π, π) (and x =

0) using |x| on the right-hand side.
14.3.2 A sawtooth wave is given by

f (x) = x, −π < x < π.
Show that

∞
(−1)n+1
f (x) = 2 sin nx.
n=1
n
14.3.3 A different sawtooth wave is described by
1
− 2 (π + x), −π ≤ x < 0
f (x) =
+ 12 (π − x), 0 < x ≤ π.
∞
Show that f (x) = n=1 (sin nx/n).
14.3.4 A triangular wave (Fig. 14.9) is represented by

x, 0≤x≤π
f (x) =
−x, −π ≤ x ≤ 0.
Represent f (x) by a Fourier series.
π 4 cos nx
ANS. f (x) = − .
2 π n=1,3,5,...
n2
Figure 14.9
f(x)
Triangular Wave
x
–4p –3p –2p –p p 2p 3p 4p
14.3.5 Expand

1, x 2 < x02
f (x) =
0, x 2 > x02
in the interval [−π, π ].
Note. This variable-width square wave is important in electronic
music.
Figure 14.10
Cross Section of
Split Tube +V
–V
14.3.6 A metal cylindrical tube of radius a is split lengthwise into two non-
touching halves. The top half is maintained at a potential +V and the
bottom half at a potential −V (Fig. 14.10). Separate the variables in
Laplace’s equation and solve for the electrostatic potential for r ≤ a.
Observe the resemblance between your solution for r = a and the
Fourier series for a square wave.
14.3.7 A metal cylinder is placed in a (previously) uniform electric field, E0 ,
with the axis of the cylinder perpendicular to that of the original field.
(a) Find the perturbed electrostatic potential.
(b) Find the induced surface charge on the cylinder as a function of
angular position.
14.3.8 Expand δ(x − t) in a Fourier series.
1 1 ∞
ANS. δ(x − t) = + (cos nx cos nt + sin nx sin nt)
2π π n=1
1 1 ∞
= + cos n(x − t).
2π π n=1
14.3.9 Verify that
1 ∞
δ(ϕ1 − ϕ2 ) = eim(ϕ1 −ϕ2 )
2π m=−∞
is a Dirac delta function by showing that it satisfies the definition of a

Dirac delta function:
π
1 ∞
f (ϕ2 ) eim(ϕ1 −ϕ2 ) dϕ2 = f (ϕ1 ).
−π 2π m=−∞
Hint. Represent f (ϕ2 ) by an exponential Fourier series.

Note. The continuum analog of this expression is developed in Sec-

tion 15.2.
14.3.10 (a) Find the Fourier series representation of

0, −π < x ≤ 0
f (x) =
x, 0 ≤ x < π.
(b) From the Fourier expansion show that
π2 1 1
= 1 + 2 + 2 + ···.
8 3 5

14.3.11 Let f (z) = ln(1 + z) = ∞ n=1 (−1)n+1 n
z /n. (This series converges to
ln(1 + z) for |z| ≤ 1, except at the point z = −1.)
(a) From the imaginary parts (item 5 of Table 14.1) show that
∞
θ cos nθ
ln 2 cos = (−1)n+1 , −π < θ < π.
2 n=1
n
(b) Using a change of variable, transform part (a) into
∞
θ cos nθ
− ln 2 sin = (−1)n+1 , 0 < θ < 2π.
2 n=1
n
14.3.12 A symmetric triangular pulse of adjustable height and width is de-

scribed by

a(1 − x/b), 0 ≤ |x| ≤ b
f (x) =
0, b ≤ |x| ≤ π.
(a) Show that the Fourier coefficients are
ab 2ab
a0 = , an = (1 − cos nb)/(nb)2 .
π π
Sum the finite Fourier series through n = 10 and through n = 100
for x/π = 0(l/9)1. Take a = 1 and b = π/2.
(b) Call a Fourier analysis subroutine (if available) to calculate the
Fourier coefficients of f (x), a 0 through a10 .
14.3.13 (a) Using a Fourier analysis subroutine, calculate the Fourier cosine
coefficients a 0 through a10 of
f (x) = [1 − (x/π )2 ]1/2 , x ∈ [−π, π ].
(b) Spot check by calculating some of the preceding coefficients by
direct numerical quadrature.
Check values. a 0 = 0.785, a 2 = 0.284.
14.3.14 A function f (x) is expanded in an exponential Fourier series
∞
f (x) = cnein x .
n=−∞
14.4 Properties of Fourier Series 683
If f (x) is real, f (x) = f ∗ (x), what restriction is imposed on the coef-

ficients cn?
14.4 Properties of Fourier Series
Convergence
Note that our Fourier series should not be expected to be uniformly convergent
if it represents a discontinuous function. A uniformly convergent series of
continuous functions (sin nx, cos nx) always yields a continuous function f (x)
(compare Section 5.5). If, however, (i) f (x) is continuous, −π ≤ x ≤ π,
(ii) f (−π) = f (+π ), and (iii) f (x) is sectionally continuous, the Fourier
series for f (x) will converge uniformly. These restrictions do not demand
that f (x) be periodic, but they will be satisfied by continuous, differentiable,
periodic functions (period of 2π ). For a proof of uniform convergence we refer
to the literature.8 With or without a discontinuity in f (x), the Fourier series
will yield convergence in the mean (Section 9.4).
If a function is square integrable, then Sturm–Liouville theory implies the
validity of the Bessel inequality [Eq. (9.73)]
∞
2 1 π 2
2a20 + a n + bn2 ≤ f (x)dx (14.42)
n=1
π −π
so that at least the sum of squares of its Fourier coefficients converges.
EXAMPLE 14.4.1 Absolutely Convergent Fourier Series If the periodic function f (x) has a
bounded second derivative f (x), then its Fourier series converges absolutely.
Note that the converse is not valid, as the full-wave rectifier (Example 14.1.2)
demonstrates.
To show this, let us integrate by parts the Fourier coefficients
π π
1 π f (x) sin nx 1
an = f (x) cos nx dx = − f (x) sin nx dx,
π −π nπ nπ −π
−π
where the integrated term vanishes. Because the first derivative f (x) is boun-
ded a fortiori, |a n| = O( n1 ), and similarly bn → 0 for n → ∞, at least as fast
as 1/n, which is not sufficient for absolute convergence. However, another
integration by parts can be done, yielding
π π
f (x) cos nx 1
an = − f (x) cos nx dx, (14.43)
n2 π n2 π −π
−π
8 See, for instance, Churchill, R. V. (1993). Fourier Series and Boundary Value Problems, 5th ed.,
Section 38. McGraw-Hill, New York.
where the integrated terms cancel each other because f is continuous and
periodic; that is, f (π ) = f (−π ). Since | f (x)| ≤ M, we find the upper bound
π
M 2M
|a n| ≤ 2 dx = 2 , (14.44)
n π −π n
which implies absolute convergence (by the integral test of Chapter 5), and
the same applies to bn. ■
If the |a n|, |bn| ≤ n−α with 0 < α ≤ 1, then we have at least conditional
convergence, and the function f (x) may have discontinuities. If α > 1, then
there is absolute convergence by the integral test of Chapter 5.
Integration
Term-by-term integration of the series
a0 ∞ ∞
f (x) = + a n cos nx + bn sin nx (14.45)
2 n=1 n=1
yields
x x x
x
a 0 x ∞
an
∞
bn

f (x)dx = + sin nx − cos nx . (14.46)
x0 2 n=1
n n=1
n
x0 x0 x0
Clearly, the effect of integration is to place an additional power of n in the

denominator of each coefficient. This results in more rapid convergence than
before. Consequently, a convergent Fourier series may always be integrated
term by term, with the resulting series converging uniformly to the integral of
the original function. Indeed, term-by-term integration may be valid even if the
original series [Eq. (14.45)] is not convergent. The function f (x) need only be
integrable. A discussion will be found in Jeffreys and Jeffreys, Section 14.06
(see Additional Reading).
Strictly speaking, Eq. (14.46) may not be a Fourier series; that is, if a 0 =
0,
there will be a term 12 a 0 x. However,
x
1
f (x)dx − a 0 x (14.47)
x0 2
will still be a Fourier series.
EXAMPLE 14.4.2 Integration of Fourier Series Consider the sawtooth series for
f (x) = x, −π < x < π. (14.48)
Comparing with Exercise 14.1.1, the Fourier series is

∞
sin nx
x=2 (−1)n+1 , −π < x < π,
n=1
n
which converges conditionally, but not absolutely, because the harmonic series
diverges. Now we integrate it and obtain item 7 of Table 14.1
x∞ ∞
sin nx (−1)n−1 x
2 (−1)n−1 dx = 2 sin nx dx
0 n=1 n n=1
n 0

∞
(−1)n−1
=2 (1 − cos nx)
n=1
n2
π2 ∞
cos nx x2
= +2 (−1)n =
6 n=1
n2 2

using Exercise 14.4.1. Because | cos nx| ≤ 1 is bounded and the series n 1/n2
converges,
x our integrated Fourier series converges absolutely to the limit
0 x dx = x 2
/2. ■
Differentiation
The situation regarding differentiation is quite different from that of integra-
tion. Here the word is caution.
EXAMPLE 14.4.3 Differentiation of Fourier Series Consider again the series of Exam-
ple 14.4.2. Differentiating term by term, we obtain

∞
1=2 (−1)n+1 cos nx, (14.49)
n=1
which is not convergent for any value of x. Warning: Check the convergence
of your derivative.
For a triangular wave (Exercise 14.3.4), in which the convergence is more
rapid (and uniform),
π 4 ∞
cos nx
f (x) = − . (14.50)
2 π n=1,odd
n2
Differentiating term by term,
4 ∞
sin nx
f (x) = , (14.51)
π n=1,odd
n
which is the Fourier expansion of a square wave

1, 0 < x < π,
f (x) = (14.52)
−1, −π < x < 0.
Inspection of Fig. 14.4 verifies that this is indeed the derivative of our trian-
gular wave. ■
• As the inverse of integration, the operation of differentiation has placed

an additional factor n in the numerator of each term. This reduces the
rate of convergence and may, as in the first case mentioned, render the
differentiated series divergent.
• In general, term-by-term differentiation is permissible under the same con-
ditions listed for uniform convergence in Chapter 5.
From the expansion of x and expansions of other powers of x, numerous

other infinite series can be evaluated. A few are included in the subsequent
exercises.
EXERCISES
14.4.1 Show that integration of the Fourier expansion of f (x) = x, −π < x <
π , leads to
π2 ∞
(−1)n+1 1 1 1
= 2
=1− + − + ···.
12 n=1
n 4 9 16
14.4.2 Parseval’s identity.

(a) Assuming that the Fourier expansion of f (x) is uniformly conver-
gent, show that

1 π a2 ∞
2
[ f (x)]2 dx = 0 + an + bn2 .
π −π 2 n=1
This is Parseval’s identity. It is actually a special case of the com-

pleteness relation [Eq. (9.73)].
(b) Given
π2 ∞
(−1)n cos nx
x2 = +4 , −π ≤ x ≤ π,
3 n=1
n2
apply Parseval’s identity to obtain ζ (4) in closed form.

(c) The condition of uniform convergence is not necessary. Show this
by applying the Parseval identity to the square wave

−1, −π < x < 0
f (x) =
1, 0<x<π
4 ∞
sin(2n − 1)x
= .
π n=1 2n − 1
14.4.3 Show that integrating the Fourier expansion of the Dirac delta function
(Exercise 14.3.8) leads to the Fourier representation of the square wave
[Eq. (14.21)], with h = 1.
Note. Integrating the constant term (1/2π ) leads to a term x/2π . What
are you going to do with this?
14.4.4 Integrate the Fourier expansion of the unit step function

0, −π < x < 0
f (x) =
x, 0 ≤ x < π.
Show that your integrated series agrees with Exercise 14.3.12.
14.4.5 In the interval [−π, π ],
δ n(x) = n, for |x| < 1/(2n),
0, for |x| > 1/(2n)
(Fig. 14.11).
(a) Expand δ n(x) as a Fourier cosine series.
(b) Show that your Fourier series agrees with a Fourier expansion of
δ(x) in the limit as n → ∞.
Figure 14.11
Rectangular Pulse d n(x)
x
–p 1 1 p
2n 2n
14.4.6 Confirm the delta function nature of your Fourier series of Exercise
14.4.5 by showing that for any f (x) that is finite in the interval [−π, π ]
and continuous at x = 0,
π
f (x)[Fourier expansion of δ∞ (x)] dx = f (0).
−π
14.4.7 Find the charge distribution over the interior surfaces of the semicircles
of Exercise 14.3.6.
Note. You obtain a divergent series and this Fourier approach fails.
Using conformal mapping techniques, we may show the charge density
to be proportional to csc θ. Does csc θ have a Fourier expansion?
14.4.8 Given

∞
sin nx −(π + x)/2, −π ≤ x < 0
ϕ1 (x) = =
n=1
n (π − x)/2 0 < x ≤ π,
show by integrating that

∞
cos nx (π + x)2 /4 − π 2 /(12), −π ≤ x ≤ 0
ϕ2 (x) ≡ =
n=1
n2 (π − x) /4 − π /(12),
2 2
0 ≤ x ≤ π.
Additional Reading
Carslaw, H. S. (1921). Introduction to the Theory of Fourier’s Series and
Integrals, 2nd ed. Macmillan, London. Paperback, 3rd ed., Dover, New
York (1952). This is a detailed and classic work.
Hamming, R. W. (1973). Numerical Methods for Scientists and Engineers, 2nd
ed. McGraw-Hill, New York. Reprinted, Dover, New York (1987).
Jeffreys, H., and Jeffreys, B. S. (1972). Methods of Mathematical Physics, 3rd
ed. Cambridge Univ. Press, Cambridge, UK.
Kufner, A., and Kadlec, J. (1971). Fourier Series. Iliffe, London. This book is a
clear account of Fourier series in the context of Hilbert space.
Lanczos, C. (1956). Applied Analysis. Prentice-Hall, Englewood Cliffs, NJ.
Reprinted, Dover, New York (1988). The book gives a well-written presen-
tation of the Lanczos convergence technique (which suppresses the Gibbs
phenomenon oscillations). This and several other topics are presented
from the point of view of a mathematician who wants useful numerical
results and not just abstract existence theorems.
Oberhettinger, F. (1973). Fourier Expansions, A Collection of Formulas. Aca-
demic Press, New York.
Zygmund, A. (1988). Trigonometric Series. Cambridge Univ. Press, Cambridge,
UK. The volume contains an extremely complete exposition, including
relatively recent results in the realm of pure mathematics.

Fourier Series PDF

Uploaded by

Copyright:

Available Formats

Fourier Series PDF

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

Fourier Series PDF

Uploaded by

Copyright:

Available Formats

Chapter 14

14.1 General Properties

Periodic phenomena involving waves [∼ sin(2π x/λ) as a crude approximation

The coefficients a 0 , a n, and bn are related to the periodic function f (x) by

Note that any interval x0 ≤ x ≤ x0 + 2π will be equally satisfactory. Fre-

1 These conditions are sufficient but not necessary.

conjugate of one of the two factors,

which is an odd function of the variable x. Hence, we expect a pure sine

which converges only conditionally, not absolutely, because of the disconti-

limited to Fourier series. It occurs with other eigenfunction expansions. For

One of the advantages of a Fourier representation over some other

f (t) = sin ωt, 0 < ωt < π,

f (x) = 0, −π < x < 0,

From Eqs. (14.2) and (14.3) we find

14.1.1 A function f (x) (quadratically integrable) is to be represented by a

Show that the requirement that

Show that this is equivalent to Eq. (14.1) with

a n = α n cos θ , α 2n = a2n + bn2 ,

Note. The coefficients αn2 as a function of n define what is called the

14.1.4 Apply the summation technique of this section to show that

14.1.5 Sum the trigonometric series

14.2 Advantages and Uses of Fourier Series

4 With the range of integration −π ≤ x ≤ π .

all bn = 0. In other words,

Frequently, these properties are helpful in expanding a given function.

f (x) = x, 0≤x<π (14.24)

a one-term series. This (one-term) series is defined for all finite x.

Fourier cosine series

• It corresponds to finding the response of the system to the funda-

f (x) = h, 0 < x < 1,

for L ≥ 1 and h > 0. The expansion interval is [−L, L].

The resulting Fourier series

and B n = 0. The Fourier series

14.2.3 In some problems it is convenient to approximate sin π x over the inter-

14.3 Complex Fourier Series

One way of summing a trigonometric Fourier series is to transform it into

Expressing cos nx and sin nx in exponential form and setting x = θ ,

u(1, θ ) + iv(1, θ ) = lim f (reiθ ). (14.37)

An application of this theorem appears as Exercise 14.3.11, and it is used in

EXAMPLE 14.3.1 Summation

Table 14.1 Fourier Series Reference

Letting r = 1 based on Abel’s theorem, we obtain item 4 of Table 14.1

14.3.1 Develop the Fourier series representation of

7 The limits may be shifted to (−π, π) (and x =

14.3.2 A sawtooth wave is given by

is a Dirac delta function by showing that it satisfies the definition of a

Hint. Represent f (ϕ2 ) by an exponential Fourier series.

Note. The continuum analog of this expression is developed in Sec-

14.3.12 A symmetric triangular pulse of adjustable height and width is de-

If f (x) is real, f (x) = f ∗ (x), what restriction is imposed on the coef-

14.4 Properties of Fourier Series

so that at least the sum of squares of its Fourier coefficients converges.

Clearly, the effect of integration is to place an additional power of n in the

f (x) = x, −π < x < π. (14.48)

Comparing with Exercise 14.1.1, the Fourier series is

Differentiating term by term,

which is the Fourier expansion of a square wave

• As the inverse of integration, the operation of differentiation has placed

From the expansion of x and expansions of other powers of x, numerous