C L ~\ ~ S I{ l) () ~\ 1 R I ~ <) l R l : I: ~I \ r I~ R I A I

J ~
 @2005 by
The Mathematical Association of America (Incorporated)
Library of Congress Catalog Control Number 2004113541
ISBN 0-88385-740-5
Printed in the United States of America
Current Printing (last digit):
10 9 8 7 6 5 ~ 3 2 I
 Fourier Series

by

Rajendra Bhatia
Indian Statistical Institute

Published and Distributed by

THE MATHEMATICAL ASSOCIATION OF AMERICA
 CLASSROOM RESOURCE MATERIALS

Classroom Resource Materials is intended to provide supplementary classroom mate1ial

for students-laboratory exercises. projects. historical infonnation. textbooks with unusual
approaches for presenting mathematical ideas. career infonnation. etc.
Council on Publications
Roger Nelsen. Chair
Zaven A. Karian. Editor
William Bauldry Stephen B Maurer
Gerald Bryce Douglas Meade
Sheldon P. Gordon Judith A. Palagallo
William J. Higgins Wayne Roberts
Paul Knopp

/OJ Careers in Mathematics, 2nd edition. edited by Andrew Sterrett

Archimedes: What Did He Do Besides Cry Eureka?. Shennan Stein
Calculus Mysteries and Thrillers, R. Grant Woods
Combinatorics: A Problem Oriellted Approach, Daniel A. Marcus
Conjecture and Proof. Miklos Laczkovich
A Course in Mathematical Modeling. Douglas Mooney and Randall Swift
Cryptological Mathematics, Robert Edward Lewand
Elemelltary Mathematical Models. Dan Kalman
En\'ironmelltal Mathematics in the Classroom. edited by B. A. Fusaro and P. C. Kenschaft
Esselltials of Mathematics: Imroduction to Theory, Proo.f. and the Professional Culture.
Margie Hale
Exploratory Ewzmples for Real Analysis, Joanne E. Snow and Kirk E. Weller
Fourier Series. Rajendra Bhatia
Geometry from Africa: Mathematical and Educational Elplorations. Paulus Gerdes
ldemijication Numbers and Check Digit Schemes, Joseph Kirtland
lllterdisciplinary Li\·ely Application Projects, edited by Chris Arney
lm•erse Problems: Acti\•itiesfor Undergraduates. C. W. Groetsch
Laboratory Elperiences in Group TheOI)', Ellen Maycock Parker
Leam from the Masters, Frank Swetz. John Fauvel. Otto Bekken. Bengt Johansson. and
Victor Katz
Math tluvugh the Ages: A Gelltle Histol)'for Teachers and Others (Expanded Edition).
William P. Berlinghoff and Fernando Q. Gouvea
Mathematical Evolutions. edited by Abe Shenitzer and John Stillwell
Mathematical Modeling in the Em•ironmellt, Charles Hadlock
Mathematics for Business Decisions Part 1: Probability and Simulation (electronic
textbook). Richard B. Thompson and Christopher G. Lamoureux
Mathematics for Business Decisions Part 2: Calculus and Optimi~ation (electronic
textbook). Richard B. Thompson and Christopher G. Lamoureux
Ordinary Differelltial Equations: A Brief Eclectic Tour. David A. Sanchez
0\'al Track and Other Permutation Pu~zles. John 0. Kiltinen
A Primer of Abstract Mathematics, Robert B. Ash
Proofs Without Words. Roger B. Nelsen
Proofs lVithow Wonls II. Roger B. Nelsen
A Radical Appmach to Real Analysis, David M. Bressoud
Resources for the Study of Real Analysis. Robert L. Brabenec
She Does fvlath!, edited by Marla Parker
Sol\'e This: lvlath Acth•ities for Studems and Clubs, James S. Tanton
Studellt lvlamwl for lvlathematics for Business Decisions Part I: Probability and
Simulation, David Williamson. Marilou Mendel. Julie Tarr. and Deborah Yoklic
Studellt lvlamwl for Mathematics for Business Decisions Part 2: Calculus and
Optimi~ation. David Williamson. Marilou Mendel. Julie Tarr. and Deborah Yoklic
Teaching Statistics Using Baseball. Jim Albert
Writing Projects for lvlathematics Courses: Crushed Clowns. Cw:~. and Coffee to Go,
Annalisa Crannell, Gavin LaRose. Thomas Ratliff. and Elyn Rykken

MAA Service Center

P. 0. Box 91112
Washington, DC 20090-1112
1-800-331-1 MAA
FAX: 1-301-206-9789
www.maa.org
Contents

Preface ix

0 A History of Fourier Series 1

1 Heat Conduction and Fourier Series 13

1.1 The Laplace equation in two dimensions 13
1.2 Solutions of the Laplace equation 15
1.3 The complete solution of the Laplace equation 19

2 Convergence of Fourier Series 27

2.1 Abel summability and Cesaro summability 28
2.2 The Dirichlet and the Fejer kemels 29
2.3 Pointwise convergence of Fourier series 35
2.4 Tenn by term integration and differentiation 44
2.5 Divergence of Fourier series 46

3 Odds and Ends 51

3.1 Sine and cosine series 51
3.2 Functions with arbitrary periods 53
3.3 Some simple examples 55
3.4 Infinite products 61
3.5 rr and infinite series 63
vii
viii FOURIER SERIES

3.6 Bernoulli numbers 65

3.7 sinxjx 67
3.8 The Gibbs phenomenon 70
3.9 Exercises 72
3.10 A historical digression 75

4 Convergence in L2 and L1 79
4.1 L'!. convergence of Fourier series 79
4.2 Fourier coefficients of L 1 functions 85

5 Some Applications 95
5.1 An ergodic theorem and number theory 95
5.2 The isoperimetric problem 98
5.3 The vibrating string 101
5.4 Band matrices 105

A A Note on Normalisation 111

B A Brief Bibliography 113

Index 117
Preface

These notes provide a quick and brief introduction to Fourier Series. The emphasis is not
only on the mathematics but also on the history of the subject, its importance, its applica-
tions and its place in the rest of science.
I first learnt about Fourier Series as a student of physics. Together with several other
assorted topics, they formed a ragbag course called Mathematical Physics from which.
when the time came, real physics courses would pick what they wanted. A little later,
as a student of mathematics I came across Fourier Series in the middle of a course on
Mathematical Analysis. On each occasion, my teachers and my books (all good) managed
to keep a secret which I learnt later. Fourier Series are not just tools for the physicist
and examples for the mathematician. They are directly responsible for the development of
nearly one half of mathematical analysis over the last two centuries.
These notes have been consciously designed to reveal this aspect of the subject and
something more. The development of Fourier Series is illustrative of a recurrent pattern in
modern science. I hope the reader will see this pattern emerge from our discussion.
This book can be used by a variety of students. Mathematics students at the third year
undergraduate level should be able to follow most of the discussion. Typically, such stu-
dents may have had their first course in Analysis (corresponding to Chapters 1-8 of Prin-
ciples of Mathematical Analysis by W. Rudin) and have a good working knowledge of
complex numbers and basic differential equations. Such students can learn about Fourier
Series from this book and, at the same time. reinforce their understanding of the analysis
topics mentioned above. More preparation is required for reading Chapter 4. a part of Sec-
tion 2.5, Sections 5.2 and 5.4. These parts presuppose familiarity with Lebesgue spaces
and elements of Functional Analysis usually taught in the fourth year of an undergraduate
or the first year of a graduate program. At many places in the book the reader will see a
statement like "Let f be a continuous 01; more generally, an integrable function." Here

ix
X FOURIER SERIES

there is a choice. If it appears easier to handle continuous functions. the reader need not be
worried about discontinuous ones at this stage.
Thus the material in this book can be used either to augment an Analysis course or to
serve as the beginning of a special course leading to more advanced topics in Hannonic
Analysis. It can also be used for a reading project. Some readers may be happy reading just
Chapter 0 outlining the history of the subject; in some sense that captures the spirit of this
book. Others may enjoy the several tidbits offered in Chapter 3.
Two editions of this book have appeared in India before this Class mom Resource Ma-
terials edition. I am much obliged to the editors of this series and to colleagues and friends
~ ~

H. Helson. A. I. Singh. S. K. Gupta. S. Serra and R. Hom for their comments and advice.
The computer drawings were made by a former student S. Guha. I am thankful to him and
to A. Shukla for preparing the electronic files.
A History of Fourier Series

A large part of mathematics has its roots in physics. Fourier series arose in the study of two
simple physical problems-the motion of a vibrating string and heat conduction in solids.
Attempts to understand what these series meant and in what sense they solved these two
problems in physics have contributed to the origin and growth of most of modern analysis.
Among ideas and theories that owe their existence to questions arising out of the study of
Fourier series are Cantor's theory of infinite sets. the Riemann and the Lebesgue integrals
and the summability of series. Even such a basic notion as that of a ''function" was made
rigorous because of these problems. A brief sketch of the history of Fourier series is given
in this chapter. You could read it now and again at the end of this course when you will
understand it better.

1. The motion of a vibrating string

It is a common experience that when an elastic string (such as a metal wire) tied at both
ends is plucked or struck it begins vibrating. For simplicity. regard the string as a one-
dimensional object occupying. when at rest. the interval 0 < x < I in the x-y plane. At
time t = 0 the string is displaced to its initial position, keeping its end points fixed, i.e .•
the shape of the string now becomes like the graph of a function

y = f(x) 0 < x < 1.

[(0) ={(I)= 0.

If the string is now released it starts vibrating. This is the model for a plucked string. We can
also consider the struck string being described as: at time t = 0 each point x is imparted
a velocity g(x) in they-direction. This is called the initial velocity. Once again the string
will vibrate. Let y = y(x. t) describe the graph of the displacement of the string at timet.

1
2 FOURIER SERIES

The motion is governed by the equations

(i) y, = a 2 yx:r· (the wm•e equation)

(ii) y(O. t) = y(l. t) = 0 for all t, (bow1da(\' conditions)
(iii) y(x. 0) = f(x). y1 (x. 0) = g(x). (initial conditions)

Here y, and Y.t denote the derivatives of y with respect tot and x. (See Section 5.3 for a
derivation of the wave equation). What is the solution of this system?
One can see that for each integer k.

y(x. t) = sinbrx sinakrrt (1)

and

y(x. t) = sinkrrx cosakrrt. (2)

both satisfy (i). They also satisfy (ii). The first satisfies (iii) if f(x) 0 and g(x) =
akrr sinkrrx. The second satisfies (iii) if f(x) = sinkrrx and g(x) = 0.
Now note that any finite linear combination of functions like (I) and (2). i.e .• a function

y(x. t) = L sin krrx(ak cosaknt + f3k sin akrr t). (3)

k

also satisfies (i) and (ii). It satisfies (iii) for certain choices off and g. But if .f and g are
any arbitrary functions. then obviously a finite sum of the form (3) is not likely to satisfy
the conditions (iii). Now notice that an infinite sum of the form (3) will also satisfy (i)
and (ii) pro\'ided term by term differentiation (~(this series is permissible. So the questions
arise: When is such differentiation permissible? If this series solution converges and can
be differentiated, can we now make it satisfy conditions (iii) for any preassigned f and g?
Are there any other solutions for our problem?
Questions like this constitute the heart of analysis. It turns out that if f and g are
sufficiently smooth. the solution we have outlined works. and is unique. (Functions in the
vibrating string problem are piecewise C 2 and that is smooth enough.) Fourier series were
introduced and developed in an attempt to answer such questions. And it is in the course of
this development that such basic concepts of mathematics as set and function were made
precise.

2. J. D'Aiembert
The wave equation that describes the motion of a vibrating string was derived by J.
D' Alembert in 1747. He gave a simple and elegant solution in the form

y(x. t) = v(at + x) - v(at - x).

which can be interpreted as a sum of two travelling waves. one moving to the left and
the other to the right. (See Section 5.3. where we derive this solution via Fourier series.
D' Alembert's solution was simpler; but the method of separation of variables we use in
these notes is more general.)
A HISTORY OF FOURIER SERIES 3

When we arc studying the motion of a plucked string: y(x. 0) = f (x ). y, (x. 0) = 0,

there is one variable f in the problem and the solution also involves one variable v which
can be determined from f. So o· AI em bert believed he had solved the problem completely.

3. L. Euler
At this time the word function had a very resuictcd meaning for mathematicians. A func-
tion meant a formula like f(x) = x 2 , f(x) = sin x, or f(x) = x tan x 2 + 1747e·'. The
formula could be complicated but it had to be a single analytic expression. Something like
f(x) = 3x 2 when 0 < x < !. and j(x) = (I - x 2 ) when ~ < x < I was not thought
of as a function on [0, 1]. Euler thought that the initial position of the string need not al-
ways be a ''function." The string could well be plucked to a shape as in Figure I and then
released. Here different parts of the string arc described by different functions. But Eu-
ler claimed that the travelling wave solution of D' AI em bert would still be valid; now the
solution would involve several different functions instead of a single function v.
In other words, at that time,fimction and graph meant two different things. Every func-
tion can be represented by its graph but not every ..graph'' that could be drawn was the
graph of a function. Euler's point was that the initial displacement of the string could be
any graph and the travelling wave solution should work here also, the two waves them-
selves being ..graphs'' now. This "physical" reasoning was rejected by D' Alembert his
"analytical'' argument worked only for functions.

FIGURE 1
Plucked String

4. D. Bernoulli
In 1755, Daniel Bernoulli gave another solution for the problem in terms of standing wa\'es.
This is best understood by considering the solution

y(x. t) = sinkJrx cosakJrt.

When k = I, the points x = 0 and x = I. i.e., the two end points of the string, remain
fixed at all times, and all other points move, the motion of any point being given as a
cosine function of time. When k = 2 the points x = 0, x = I and also the point x = I 12
remain fixed at all times: the rest of the points all move; at any fixed time the string has
the shape of a sine wave; and the motion of any point is given by a cosine function of
time. The point x = I 12 is called a node. For an arbitrary k > I. the end points and the
points I 1k, 21 k, .... (k - I) I k on the string remain fixed while all other points move as
described. These are called ..standing waves"; the (interior) fixed points are called nodes
and the motion for k = I, 2.... is called the ji1:.:;t harmonic, the second harmonic and so
on. Bernoulli asserted that every solution to the problem of the plucked string (y(x. 0) =
j(x), y 1 (x, 0) = 0) is a sum of these harmonics, and the sum could be infinite.
4 FOURIER SERIES

Euler strongly objected to Bernoulli's claim. His objections were on two grounds. First
of all, Bernoulli ·s claim would imply that WI)' function f (x) could be represented as

(4)

because at timet = 0 the initial position could be any {(x). However, the right-hand side
of (4) is a periodic function. whereas the left-hand side is completely arbitrary: further the
right-hand side is an odd function of x whereas the left-hand side is completely arbitrary.
Second. this would not, in any case, be a general solution to the problem. The right-hand
side of (4) is an analytic fonnula. even though an infinite series. hence is a function. How-
ever, as he had pointed out earlier in connection with D' Alembert's solution, the initial
position of the string could be any graph but not necessarily a function. So, Euler believed
that D' Alembert's travelling wave solution was applicable to the general case of the ini-
tial position being a graph-which D'Alembert himself did not believe-but Bernoulli's
solution would be applicable only to functions and to very restricted ones.
Bernoulli did not contest Euler's point about functions and graphs. However. he insisted
that his solution was valid for all functions. His answer to Euler's first objection was that
the series (4) involved infinitely many coefficients O:k and by choosing them properly one
could make the series take the value f (x) for infinitely many x. Of course, now we know
that if two functions f and g are equal at infinitely many points, then they need not be
equal at all points. However, at that time the nature of infinity and of a function was still
not clearly understood and so this argument sounded plausible. Euler did not accept it. but
for different reasons.
The series (4) is now known as a Fourier series and the coefficients O:k are called the
Fourier coefficients of f. Now we know that not every continuous function can be written
in this fonn, but still Bernoulli was almost right, as we will see. Assuming that f can be
represented as (4) and that the series can be integrated tenn by tenn. the coefficients O:k are
easily seen to be

ak = 21 1
j(x) sinklrx dx. (5)

Euler derived this fonnula by a complicated argument and then noticed it was an easy
consequence of the orthogonality relations

1 1
sinbrx sin11mxdx = 0. k=ftm. (6)

5. J. Fourier
In 1804, Joseph Fomier began his studies of the conduction of heat in solids and in three
remarkably productive years discovered the basic equations of heat conduction. developed
new methods to solve them. used his methods to analyse several practical problems and
supplied experimental evidence to support his theory. His work was described in his book
Tlte Analytical Theory of Heat, one of the most imp011ant books in the history of physics.
Let us describe one of the simplest situations to which Fourier's analysis can be applied.
Consider a 2-dimensional disk. say the unit disk in the plane. Suppose the temperature at
A HISTORY OF FOURIER SERIES 5

each point of the boundary of the disk is known. Can we then find the temperature at any
point inside the disk? At steady state the temperature u(r. 0) at a point (r. 0) obeys the
equation

(i) (ntr)r + lfr uoo = 0.

This is called the Laplace equation in polar fonn. (See Section 1.1 for the derivation of
this equation from basic principles.) We are given the temperatures at the boundary, i.e .•
we know

(ii) u( I. 0) = /(8).
where f is a given continuous function. The problem is to find u(r. 0) for all (r. 0). This is
the same kind of problem as we considered earlier for the vibrations of a string. Using an
analysis very much like that of Bernoulli, Fourier observed that any finite sum
N
u(r. 0) = L A,rl"lei"O
11=-N

is a solution of (i). Of course. such a sum will not satisfy the boundary condition (ii) unless
f is of a special type. Fourier asserted that an infinite sum
00

u(r. 0) = L A,rl"lei"O
11=-x

is also a solution of (i). and further by choosing A 11 properly it can be made to satisfy the
boundary condition (ii). In other words we can write
00

f(O) = L A,ei"O {7)

11=-X

when f is any continuous function.

So far, this is parallel to Bernoulli's analysis of the vibrating string. But Fourier went
a step further and claimed that his method will work not only for f given by a single
analytical formula but for f given by any graph. In other words. now there was to be no
distinction between a function and a graph. Indeed, if Fourier's claim was valid, then every
graph would also have a formula, namely the series associated with it.
Fourier, like Euler, calculated the coefficients A, occurring in {7). by a laborious (and
wrong) method. These are given by

1
A,=- jrr fW)e-i" dO. 0
(8)
2Jr -Tr

The series {7) is now also called a Fourier series and the coefficients (8) the Fourier co-
efficients of f. Fourier noticed that the coefficients A, are meaningful whenever f is a
graph bounding a definite area (integrable in present day terminology). So he claimed his
solution was valid for all such f.
6 FOURIER SERIES

Fourier's theory was criticised by Laplace and Lagrange. among others. However they
recognised the importance of his work and awarded him a major prize.
Since einO =cos nO+ i sin nO. the series (8) can be rearranged and written as
00
0
J(O) = ~ + L<all cosne + bll sin nO). (9)
- 11=1

This too is called the Fourier series for f.

Now, what about Euler's two objections to Bernoulli's solution? (Euler died in 1783
and was not there to react to Fourier's work.) How could an arbitrary function be a sum
of periodic functions? The way out of this difficulty is astonishingly simple. The given
function is defined on some bounded intervul, say [0. I] or [0. rr ]. where it represents some
physical quantity of interest to us,like the displacement of a string. We can extend it outside
this interval and make the extension periodic. and either odd or even. For example, if
f(x) = x. 0 < x < 1. is the function given to us. we can define an extension of it by putting
f(x) = -x. -I < x < 0, and then further extending it by putting f(x + 2k) = j(x).
-I < x < I. k E Z. This defines an even function of period 2. Its graph is shown
in
.
F1gure ?
-·

FIGURE 2
Even periodic extension of f

We could also have extended f from its original definition .f(.r) = x. 0 < x < I. by
putting f(x) = x. -1 < x < 0 and then extending it further by putting f(x + 2k) =
f(x). -1 < x < I. as before. Now we get an odd function of period 2: the graph of this
function is shown in Figure 3. The Fourier series in the first case is
1 4 4
f(x) = - -----:; cosrrx - -..,cos 3rrx- ...
2 rr- 9rr-

FIGURE 3
Odd periodic extension of f
A HISTORY OF FOURIER SERIES 7

and in the second case it is

") ") ")

f(x) =.:. sinJTx- _:_ sin:2rrx + _:_ sin3rrx- · · ·

JT 2JT 3rr
(These expansions. and several others are derived in Section 3.3). In the interval [0, I) both
se1ies converge to the same limit. Notice that the terms of one series are all even functions
and of the other are all odd. Of course if a function is thought of as a formula. then these
extensions are not functions. Fourier's idea of admitting functions which arc identical on
some interval but different elsewhere made it possible to <tpply his theory to a wide variety
of situations. It also led to a critical examination of the notion of a function itself.
As Fourier observed, the coefficients A, in (8) made sense for all functions for which
the integral is finite. He believed that the series (7) would equal f for all such .f at all fJ.
Here he was not right, but in a nontrivial sense. For all good (piecewise smooth) functions
which occuned in his problems he was right.

6. P. Dirichlet
Fourier did not state or prove any statement about Fourier series that would be judged to
be ··conect.. in a mathematics examination today. It was Dirichlet who took up Fourier"s
work and made it into rigorous mathematics. In the process he laid firm foundations for
modern analysis.
First of all it was necessary to have a clear definition of a function. Dirichlet gave the
definition which we learn now in our courses: a function is a rule which assigns a definite
value .f(x) to any x in a certain set of points. Notice now that a function need no longer
be a ..graph;· let alone a formula. In fact Dirichlet in 1828 gave an example that Fourier
could not have imagined. This is the characteiistic function of the set of rational numbers:
.f (x) = I if x is rational and .f (x) = 0 if x is im:ttional. This function cannot be represented
by any ··graph:· So now The subjecT of analysis was no longer a pan of geometry. Notice
also that this function does not .. bound any area:· so its Fourier coefficients cannot be
calculated by Fourier's methods.
However, for all functions .f that can be ..drawn:· i.e .• for piecewise smooth functions .f.
Dirichletpro,·ed that the Fourier series off converges to .f(x) at every point x where f is
continuous, and to the average value !<.f<x+) + .f(.·L)) if .f has a jump at x. Further, if .f
is smooth on an interval [a. b]. then its Fourier series converges un{formly to f on [a, b].
This was the first major convergence result for Fourier series. A later example is C. Jordan ·s
theorem that says that continuous functions of bounded variation have convergent Fourier
series. Dirichlet's theorem was a forerunner of several others giving sufficient conditions
for a function to have a convergent Fourier series. (Some of these theorems are proved in
Chapter 2. Section 3.)

7. B. Riemann
To handle functions that are not graphs. i.e., those that have more than finitely many turns
and jumps. one would need to generalise the notion of an integral beyond the intuitive idea
8 FOURIER SERIES

of the ··area under a curve." Only then can we hope to calculate the Fourier coefficients of a
function with infinitely many discontinuities. Riemann developed his theory of integration.
which could handle such functions-that is the integral that we learn first now.
Using this integral Riemann gave an example of a function that does not satisfy Dirich-
let's or Jordan's condition but has a pointwise convergent Fourier series. Riemann initiated
the study of trigonometric series that need not be Fourier series of any function.

8. P. du Bois-Reymond
Dirichlet believed (and thought that he would soon prove) that the Fourier series of every
continuous function, and perhaps of every Riemann integrable function, converges at every
point. This belief came to be shared by other prominent mathematicians like Riemann,
Weierstrass and Dedekind. In 1876, however, Du Bois-Reymond proved them wrong by
constructing an example of a continuous function whose Fourier series is divergent at one
pomt.
You might have come across the Weierstrass construction of an example of a continu-
ous function which is not differentiable at any point. The function is obtained by succes-
sively constructing worse and worse functions. Du Bois-Reymond's example is a similar
construction. This method is called the principle of condensation of singularities. (The
construction of the example and the general method are described in Chapter 2, Section 5.)

9. G. Cantor
Cantor's theory of sets and infinite numbers is now the basis for all analysis. This theory
too owes its existence, at least in part, to Cantor's interest in Fourier series. He observed
that changing a function f at a few points does not change its Fourier coefficients. So the
behaviour off at a few points does not matter for Fourier analysis. How many points can
be ignored in this way and what kind of sets do they constitute? This problem led Cantor
to his study of infinite sets and cardinal numbers.

10. L Fejer
Consider the series L x, where x, = ( -1 )", 11 > 0. The partial sums s, of this series are
alternately I and 0. So the series does not converge. However, its divergence is different
in character from that of the series L n or L 1/n. In these latter cases the partial sums
diverge to oo, whereas in the former case the partial sums oscillate between I and 0 and
hence on an average take the value I/2. We say that a series L x, is (C, I) sum mabie
or Ces(uv summab/e if the sequence a, = ~(so + s 1 + .. · + s,_ 1). formed by taking
the averages of the first n partial sums of the series, converges. Every convergent series is
(C, I) sununable~ a series may be (C, I) summable but not convergent. for example this is
the case when x, = (- I )11 •
In 1904. Fejer proved that the Fourier series of every continuous function is (C. I)
summable and, in this new sense, converges uniformly to the function. This theorem is ex-
tremely useful and gave an impetus to the study of sununability of series. (Fejer's Theorem
is proved in Chapter 2, Section 2.)
A HISTORY OF FOURIER SERIES 9

11. H. lebesgue
It had already been noted, by Cantor among others, that changing a function at ··a few"
points does not alter its Fourier series, because the value of the integral (8) defining the
Fourier coefficients is not affected. So it is not proper to ask whether the sum of the Fourier
series of f is equal to f at e\'et)' point; rather we should ask whether the two are equal
everywhere except on those sets that are irrelevant in integration.
This problem led to a critical examination of the Riemann integral. Just as Riemann was
motivated by problems in Fourier series to define a new concept of integration, the same
motivation led Lebesgue to define a new integral that is more flexible. The notions of sets
of measure zero and al11wst e\'erywhere equality offimctions now changed the meaning of
function even more. We regard two functions f and g as identical if they differ only on a
set of measure zero. Thus, Dirichlet's function (the characteristic function of the rationals)
is equal to 0 ··almost everywhere." So, from a formula to a graph to a rule, a function now
became an equi,•alence class.
The Lebesgue integral is indispensable in analysis. Many basic spaces of functions are
defined using this integral. We will talk about the L 11 spaces

LP = {f : j I!I oo} . I
P < < p < oo,
in particular.

12. A.N. Kolmogorov

Before Du Bois-Reymond's example in 1876, mathematicians believed that the Fourier
series of a continuous function will converge at every point. Since they had failed to prove
tllis, the example made them think that perhaps the very opposite might be true-that there
might exist a continuous function whose Fourier series diverges at e\'ery point.
In 1926 Kolmogorov proved something less but still very striking. He proved that there
exists a Lebesgue integrable function defined on [-rr. 1T ]. i.e., a function in the space
L 1([ -JT, JT ]), whose Fourier series diverges at every point.

13. l. Carleson
The function constructed by Kolmogorov is not continuous. not even Riemann integrable.
However, after his example was published it was expected that sooner or later a continuous
function with an everywhere divergent Fourier series would be discovered. There was a
surprise once again. In 1966 Carleson proved that iff is in the space L2([ -JT, JT]), then its
Fourier series converges to f at almost all points. In particular. this is true for continuous
functions. So Fourier had been almost right!
Carleson 's theorem is much harder to prove than any of the other results that we have
mentioned. This whole investigation. which had occupied some of the very best mathe-
maticians over two centuries. was nicely rounded off when in 1967, R.A. Hunt proved that
the Fourier series of every function in L,, 1 < p < oo. converges almost everywhere; and
in the converse direction Y. Katznelson and J.P. Kahane proved in 1966 that given a set E
10 FOURIER SERIES

of Lebesgue measure 0 in [-rr. rr ]. there exists a continuous function whose Fourier series
diverges on E.

14. The L2 theory and Hilbert spaces

We have already spoken of the space L2 , and earlier of orthogonality of trigonometric
functions. The space L 2 is an example of a Hilbert space; in such spaces an abstract no-
tion of distance and of orthogonality arc defined. The functions e, (x) = ei"x I -Jfii. n =
0. ±I, ±2 •.... constitute an orthonormal basis in the space L2 ([ -rr, rr ]). So the Fourier
series off is now just an expansion off with respect to this basis: f (x) = L~oo A, e, (x):
just as in the three-dimensional space JR3 • a vector can be written as a sum of its compo-
nents in the three directions. The infinite expansion now is convergent in the metric of the
space L2. i.e.,

lim
N-'X:J
(ln
-rr
J(x)- t
n=- N
A.,e.,(x)
2

dx) = 0.

This notion of convergence is different from that of convergence at every point. or at almost
every point; but, in a sense. it is more natural because of the interpretation of the Fourier
series as an orthogonal expansion. (We study this in Chapter 4.)

15. Some modern developments-1

One of the basic theorems in functional analysis says that the Hilbert spaces L2([ -rr, rr])
and /2 are isomorphic. This theorem, called the Riesz-Fischer Theorem, is proved using
Fourier series. It is taken as the starting point of J. von Neumann's .. transfommtion the-
ory'' which he developed to show the equivalence of the two basic approaches to quantum
mechanics called matrix mechanics and Wa\le mechanics.
Another .. transfonnation theory" achieving the same objective was developed by P.A.M.
Dirac. In this theory a crucial role is played by the ··iS-function." This has some unusual
properties not consistent with classical analysis. (In von Neumann's book Mathematical
Foundations of Quantum Mechanics the iS-function is described as •·fiction.") Among other
things the iS-function satisfies the properties tS(x) = tS( -x), tS(x) = 0 for all x except for
X = 0 and r D(x) = 1. Clearly now the integral could neither mean the area under a graph
nor the Riemann integral nor the Lebesgue integral. The study of such objects, which are
now called generalised functions or distributions. is a major branch of analysis developed
by S.L. Sobolev and L. Schwartz. In some sense, spaces of such functions have become
the most natural domain for Fourier analysis.

16. Some modern developments-11

We have talked only of Fourier series. A related object is the Fourier integral introduced
by Fourier to study heat conduction in an infinite rod. This is the integral defined as
A HISTORY OF FOURIER SERIES 11

j(t) = i: J<x)e_;,_, dx, t E ffi:

and is called the Fourier transform of the function f. Notice the similarity between this and
the Fourier coefficients where the v::u·iable t is replaced by the integer n. and the domain of
integration is [-JT,JT] instead of (-oo. oo).
On the one hand this object has been used in such ureas as heat conduction. optics. signal
processing. and probability. On the other hund the subject has been made more abstract
in a branch of mathematics called harmonic analysis. Integers and the interval [-JT.JT)
identified with the unit circle both form groups. Fourier series constitute harmonic analysis
on these groups. The Fourier transfom1 belongs to hannonic analysis on the group 1ft A
very similar analysis can be done on several other groups.
Major contributions to the "applied" branch were made by N. Wiener who developed
what he called ··generalised harmonic analysis." "Pure.. harmonic analysis has developed
into one of the central meas in mathematics. Some of the major figures in its development
were E. Cmtan. H. Weyland Harish-Chandra. And this brings us to our final highlight.

17. Pure and applied mathematics

Fourier series were invented by Fourier who was studying a physical problem. It is no won-
der then that they have applications. Of course not all creatures of "applied.. mathematics
have applications in as wide an area as Fourier series do. As this account shows. attempts
to understand the behaviour of these series also laid down the foundations of rigorous ... anal-
ysis. Questions like uniform convergence. Cesaro summability and subjects like transfinite
cardinals and Lebesgue measure are thought of as "pure" mathematics. Their history too is
related to Fourier series.
Among the purest branches of mathematics is number theory-and surely it is a subject
quite independent of Fourier series. Yet Heimann Weyl used Fejer's convergence theorem
for Fourier series to prove a beautiful theorem in number theory. called the Weyl Equidis-
tribution Theorem. Every real number x can be written as x = [x] + .r. where [x J is the
integral part of x and is an integer..r is the fractional part of x and is a real number lying
in the inten,al [0. I). Weyl's theorem says that if x is an irrational number then. for large N,
the fractional parts .r. (2x)..,. .... , (N x)..,. are scattered unifonnly over (0, I). (This theorem
is proved in Chapter 5. Section I.)
One of the areas where Fourier series and transfonns have major applications. is crystal-
lography. In 1985 the Nobel Prize in Chemistry was given to H. A. Hauptman and J. Karle
who developed a new method for calculating some crystallographic constants from their
Fourier coefficients, which can be inferred from measurements. Two crucial ingredients
of their analysis are Weyl's equidistribution theorem and theorems of Toeplitz on Fourier
series of nonnegative
... functions .
Fast computers and computations have changed human life in the last few decades. One
of the major tools in these computations is the Fast Fourier Transform introduced in 1965
by J. Cooley and J. Tukey in a sh01t paper titled An algorithm for the machine calculation
of complex Fourier series. Their idea reduced the number of arithmetic operations required
in calculating a discretised version of the Fourier transform from O<N'!..) to O(N log N).
12 FOURIER SERIES

This went a long way in making many large calculations a practical job. (Note that if
N = I 03, then N 2 = I 06 but N log N is approximately 7000.) In 1993 it was estimated
that nearly half of all supercomputer central processing unit time was used in calculating
the Fast Fourier Transform-used even for ordinary multiplication of large numbers.
Another major advance in the last few years is the introduction of wm'elets. Here func-
tions arc expanded not in terms of Fourier series, but in terms of some other orthonormal
bases that are suited to faster computations. This has led to new algorithms for signal pro-
cessing and for numerical solutions of equations.
If conduction of heat is related to the theory of numbers, and if theorems about numbers
are found useful in cherrustry, this story has a moraL The boundary between deep and
shallow may be sharper than that between pure and applied.
 • •• • • • /, :$. •• J' : • • --J • • ; ...... 1. ....

·~·~ ' ' '':(I :v ' •. .~' :\,.;~,_.,~ r~ ~ :f. I

Heat Conduction and Fourier Series

In this chapter we derive the mathematical equations that describe the phenomenon of heat
flow in a thin plate. We explain how, and in what sense, these equations are solved using
Fourier Series. Our analysis leads to several questions. Some of these are answered in
Chapters I and 2.

1.1 The laplace equation in two dimensions

The equation

a2u a2u
-+ -- 0 ( 1.1)
ax
2 2 ay -

is called the 2-vruiable Laplace equation. Here u(x. y) is a function on IR2 which is of
class C 2, i.e., u has continuous derivatives up to the second order.
This equation describes natural phenomena such as steady flow of heat in two dimen-
sions. To understand this imagine a thin plate of some material. We can view this as a
section of the plane with some boundary. If different parts of the plate are at different tem-
peratures, then heat flows from a point at higher temperature to one at lower temperature
r
according to Newton's /au· of cooling. This says that if is a curve in the plane, then the
rate at which heat flows across r is equal to k fr<aujan)dy. where u(x. y) is the tem-
perature at the point (x. y) of the plate, au ;an denotes the derivative of It with respect to
r
arc length along any curve normal to and k is a constant called the thermal conducth'ity
of the material of which the plate is made. (This is an experimental fact which describes
the situation very accurately for moderate temperature gradients. We will accept this as a
postulate.)
Now consider a small rectangular section of the plate

{(x, y): xo < x < x,, )'o < y < yl}.

13
14 FOURIER SERIES

If the plate is in thermal equilibrium, i.e., there is a steady state in which there is no net heat
flow into this small section of the plate or out of it, then calculating the heat flow across
the four boundaries of the above section according to Newton's law and equating it to 0
(steady state) we get

!. la,~
YI
Yo a.,
(XI, y) -
a., xo
l
a,~ (xo. y)} dy + 1XI a,~ (x . .\'I) -
a-'
:Ity (x. Yo)} dx = 0.

The negative signs here correspond to our understanding that if heat is flowing into the
rectangle from one side, then it is flowing out of it from the opposite side. Divide this
equation by XI - xo and then let x1 ~ xo. One gets
}"1 a2u au au
-., (xo, y) dy + -(xo, Yl)- -(xo, Yo)= 0.
YO
/. ax- ay ay
Now divide by .\'I -Yo and let Yl ~ YO· This gives

a2u a2u
-., (xo. Yo) + -., (xo, Yo) = 0.
ax- ay-
This is the Laplace Equation ( 1.1) and as we have seen above it describes the phenomenon
of steady state heat conduction in two dimensions.
We will find it more convenient to use polar coordinates. In these coordinates ( 1.1) can
be transformed to
a2 u I au I a2 u
-+ - - + - - -= 0 ( 1.2)
ar 2 ,. 2ar 2 r ao
for r # 0. where as usual (r. 0) are the polar coordinates of a point in the plane.

Exercise 1.1.1. Derive ( 1.2) from ( 1.1) by a change of variables. Hint: Use the relations
au au I au
-=cosO--- sinO-
ax ar ,. ao'
au au I au
-=sinO-+- cosO-.
ay ar r ao
which can be derived from the basic relations X = /" COS 0, y = r sin 0.

Exercise 1.1.2. Derive ( 1.2) directly from Newton's law. Hints: Instead of the rectangular
element that was used earlier consider now the element

{(r. 0) : ro < r < 1"1 • Oo < 0 < 0 d.

To calculate the rate of heat flow across the side 0 = 00 • choose as curves normal to it the
circular curves r = constant. Then note that the nonnal derivative with respect to the arc
length along these curves is

au . u (r. Oo + tSO) - u (r. Oo)

an Inn
ho--o rtSO
= -rI -au
( r 6)
ao ' 0 .
Add the contributions from the four sides of the element and proceed as before.
HEAT CONDUCTION AND FOURIER SERIES 15

Now consider a circular disk

D = {(r. 0) : 0 < r < I. -rr < 0 < rr}.

As usual the points (r. -rr) and (r, rr) arc considered to be identical. At steady state the
temperature u(r. 0) for all points 0 < r < I on D satisfies equation ( 1.2). Extend this to
the origin by the natural condition

u(r. 0) is continuous on D. ( 1.3)

We must also have

u(r. 0) = u(r. 0 + 2rr ). ( 1.4)

By a bmmdw)' condition we mean a condition of the type

u(l, 0) = f(B). ( 1.5)

where f is a given function defined on [-rr, rr] that is continuous and for which f( -rr) =
f(rr).
The problem of finding a function u satisfying ( 1.2)-( 1.5) is called a Dirichlet problem.
This is a typical instance of a bmmdw)' value problem in physics. In the present situation
this is: the temperature at the boundary of D is known to us and is given by ( 1.5)~ assuming
that the temperature function inside D satisfies ( 1.2)-( 1.4) can we find it? In the remaining
part of this chapter we shall answer this question.

Exercise 1.1.3. Note that every constant function satisfies ( 1.2)-( 1.4). Show that the func-
tion u(r, 0) = logr satisfies ( 1.2) and (1.4) but not (1.3).

1.2 Solutions of the laplace equation

Let us first derive solutions of two ordinary differential equations.
The first is the second order differential equation with constant coefficients:

)' + ay + by = 0.
II I ( 1.6)

The familiar method of solving this is the following. If L denotes the differential operator

d2 d
L=-+a-+b.
dx 2 dx
then ( 1.6) can be written as

L(y) = 0. ( 1.7)

Note that
16 FOURIER SERIES

and hence y = eax is a solution of (I. 7) provided a is a root of the polynomial equation
.,
p(a) =a-+ aa + b = 0. (1.8)

This is called the characteristic equation associated with the differential equation ( 1.6). If
this quadratic equation has two distinct roots a1 and a2. then the solutions eap· and ea'!x
of ( 1.6) are linearly independent and hence their linear combination

( 1.9)

gives the general solution of ( 1.6). If ( 1.8) has a double root ao. then one solution of ( 1.6)
is eaox. In this case p(ao) = p'<ao) = 0 and one can see that

So. in this case xeao-'" is another solution and the most general solution of ( 1.6) is of the
type

( 1.10)

The second equation that will arise in our analysis is the equation
'l II I
.cy +axy +by= 0. x>O (1.11)

which is called an equation of Cauchy type, in which the coefficient of y(") is x". Proceed-
ing as earlier. let

., d 2 d
L = ;c -., +ax - +b.
dx- dx
and note

where.

p(a) = a(a - I)+ aa +b.

Hence y = xa is a solution of ( 1.11) provided a is a solution of the characteristic equation

p(a) = 0.
So. if this equation has two distinct roots a, and a2. then the general solution of ( 1.11) is
( 1.12)

In case p(a) = 0 has a double root ao. then one can check that xau and .ran logx arc two
linearly independent solutions of ( 1.11) and hence the general solution. in this case, is

(1.13)
HEAT CONDUCTION AND FOURIER SERIES 17

Now the Laplace equation can be solved by a standard device-the separation method
which can (sometimes) reduce a partial differential equation to ordinary differential equa-
tions. This proceeds as follows. Let us assume, tentatively, that ( 1.2) has a solution of the
fonn

u(r, 0) = R(r)F(O), (1.14)

where R and F are functions of r and 0 alone, respectively. Then ( 1.2) becomes

r 2 R F + r R' F + R F = 0.
11 11

This is a mixture of two ordinary differential equations, which can be .. separated" if we

divide by R F and then rearrange tenns as

., R 11 R' F 11
r--+r- = - - .
R R F
Notice that now one side does not depend upon 0 and the other side docs not depend
upon r. Hence the quantity on either side of this equation must be a constant. Denoting this
constant by c, we get two equations

F II + cF = 0. ( 1.15)
.,
,--R II
+ rR I - c R = 0, (1.16)

which are ordinary differential equations of the type ( 1.6) and ( 1.11 ), respectively.

Exercise 1.2.1. Show that the equations ( 1.15) and ( 1.16) have the following solutions:
(i) if c > 0. then

F(O) = Aei../CO + Be-i ..;co,

R(r) = ar..;c + br-../C;
(ii) if c = 0. then

F(O) =A+ BO.

R(r) =a +blogr;

(iii) if c < 0, then

F(O) = AeFco + Be-Fc0 •

R(r) = ariFc + b,--iFc.
Here A, B, a. b are constants.
(Though our function u(r, 0) is real, we will find it convenient to use complex quantities.
If r is positive and a is any real number, then we have ria = eia logr = cos(a log r) +
i sin(a log r)).
 FOURIER SERIES
18

In each of the three cases above u = R F gives a solution of the Laplace equation ( 1.2).
However. our other conditions rule out some of these solutions. The condition ( 1.4) says
that u (r. e) is a periodic function of e with period 2rr. The function u = R F in case (iii)
of Exercise 1.2.1 fails to meet this requirement. In case (ii) the function F(B} is periodic
only when B = 0. and the function R(r) is bounded only when b = 0. Thus the only
admissible solution in case (ii) is u = A a. In case (i) the condition (1.4) is satisfied iff
c = n2. n = I. 2..... With these values of c. R is bounded at the origin if b = 0. So the
solution in Case (i) has the fonn u = ar" (Ae;,e + Be-i" ). n = I. 2.....
0

Thus, we have shown that all solutions of ( 1.2)-( 1.4) that are of the fom1 ( 1.14) are
given by

ll = 0, I. 2....

where a. A. B are constants. Now notice that (finite) linear combinations of solutions of
( 1.2)-( 1.4) are again solutions. So the sums

N
L A,rl"lei"tJ. (1.17)
11=-N

where A, are constants all satisfy ( 1.2)-( 1.4) for N = I, 2, ....

The constants A, will be determined from the boundary condition ( 1.5 ). However it is
clear that we cannot expect an arbitrary function f to be equal to a sum L~=-N A,ei 0.
11

One may now wonder whether an infinite sum

L Allrl"liiiiJ
tX>
(1.18)
11=-0::,

is also a solution of ( 1.2)-( 1.4). If so. then can the boundary condition ( 1.5) be satisfied by
choosing the coefllcicnts A 11 properly. i.e., do we have
'X)

{(B) = u( I. B)= L A,ei" 0

. ( 1.19)
11=-CX.

Fourier asse1ted that this is indeed so for (viituallyl every function f. He was not quite
right. However. his analysis of this problem led to several developments in mathematics
which we mentioned at the beginning.
To sum up: our analysis till ( 1.17) has been quite rigorous. Now we would like to answer
the following questions:
I. For what functions f can we choose constants A, so that the equation ( 1.19) is satisfied?
If this equation is not satisfied in the usual sense of .. pointwise convergence" is it true
in some other sense? How arc the coefficients A 11 dctcnnined from f?
2. If A, are chosen properly then is ( 1.18) indeed a solution of the Dirichlet problem?

Exercise 1.2.2. Some mystery may be taken out of the above problem if you use your
knowledge of complex analysis. Any function u(x. y) satisfying ( 1.1) is called a harmonic
HEAT CONDUCTION AND FOURIER SERIES 19

function. A function that is harmonic in the region D is the real part of a complex analytic
function, i.e., there exists an analytic function g(::) on D such that g = 11 + iv. Every such
g has a power series expansion L~oa 11 ~ 11 convergent in the interior of D. The coefficients
a" can be determined from g by Cauchy's integral formula. Recall that this power series
does not converge at all points on the boundary of D. Usc these facts to answer questions
about Fourier series raised above.

1.3 The complete solution of the laplace equation

We denote by T the boundary of the disk D. We can identify T with the interval [-11, 11],
where the points -11 and 11 arc regarded as the same. By a function on T we mean a
function on [-11, 11] such that f (-11) = f (11). Such a function can also be thought of as
representing a periodic function on IR with period 211. So, we will use the tenns .. function
on T'' and .. periodic function on IR with period 211" interchangeably.
Let f be a continuous function on T. We want to know whether it is possible to write f
as an infinite series
00
[(0) = E AneinO ( 1.20)
11=-00

for some coefficients A 11 depending on f. Suppose this is possible. Then what should A 11

be? Recall that

if Ill # 0,
if Ill= 0.

So, if we do have an expansion like ( 1.20) and tf further this series could be integrated
term by term, then we must have

An = - I ~;r f(O)e-inu dO. (1.21)

211 -;r

So given a continuous, or more generally, an integrable function f on T let us define

for each integer n,

j(n) = _I ~rr f(O)e-inO dO. ( 1.22)

211 -rr

Each j(n) is well defined. This is called the nth Fourier coefficient of f. The series
'XI
E j(n)einO ( 1.23)
11=-00

is called the Fo11rier series of f At this moment we do not know whether this series
converges, and if it converges whether at each point 0 it equals /(0).
Now let us return to the Dirichlet problem. Consider the series ( 1.18) where An are
given by ( 1.21 ); we get a series
20 FOURIER SERIES

( 1.24)

Now for r < I the series L~oo rl 11 leiii(O-I) converges unifonnly in t. (Use the M-test).
Hence the sum and the integral in ( 1.24) can be interchanged. Also, using the identities

oc I
Lxll = - -..
II= 0
I-·'

-1 X· X
L
11=-00
x -II = L xll =
11=1
I - x.

both valid when lx I < I, one obtains

oo I ,.2
"'"""',-1 11 li 11 <0 -l) = - ., for r < I.
~
-:-.... I - 2r cos(O - t) + r-

Thus the series ( 1.24) converges for all r and 0. when 0 < r < I and -rr < 0 < rr.
Call the sum of this series ll(r, 0); we have

ll(r.O) =-
2rr
I ln -rr I - 2r cos(O-
I - ,.2
t) + r~
.,J(t)dt. ( 1.25)

This is called the Poisson integral of f. The function

..,
I 1-r-
P(r. 0) = - . , . 0 < r < I. ( 1.26)
2rr I - 2r cos 0 + r- -

is called the Poisson kernel.

A simple calculation shows that

2 2
a 2 a a
(r -ar 2 + rar- +-
)
(r" cos nO)= 0. for alln.
ao 2

Since for r < I the series I::x rl 11 lei 11 H. being absolutely and unifonnly convergent. can
be differentiated tem1 by term. this shows that ll(r. 8) as dcl1ned above is a solution of the
Laplace equation. Thus ll(r,O) defined by (1.25) satisfies (1.2). (1.3) and (1.4). We will
show that it also ..satisfies'' the boundary condition ( 1.5) in the sense that ll(r. 0) tends to
{(0) uniformly as r - I.
Let f and g be two periodic functions on~ with period 2rr. Iff and g arc continuous.
we define their com·olution .f * g by

<f*g)(x) = ~~ f<x -t)g(t)dt.

HEAT CONDUCTION AND FOURIER SERIES 21

More generally, iff and g are two integrable functions, then

~~ IJ<x - l)g(l)l dt < oo

for almost all x. For these x we define (f * g)(x) = f~rr f(x- t)g(t) dt.
It is easy to see that f * g = g * f.
Convolution is thus a commutative binary operation on the space of continuous periodic
functions. Show that this operation is associative.
Does there exist an ..identity'' element for this operation? There is no continuous func-
tion g such that g * f = f for all f. (See Exercise 4.2.5.) However. there is a modified
notion of ..approximate identity" in this context which is useful. We can find a family of
functions Q, such that Q, * f converges to f uniformly on [-rr. rr] for all continuous
functions f.
We call a sequence of functions Q, on [-rr. rr] a Dirac sequence if

(i) Q, {t) > 0. (1.27)

r:
(ii) Q, (-t) = Q, (t), (1.28)

(iii) Q.. (I) dt = ~. (1.29)

(iv) for each e > 0 and D > 0, there exists N such that for alln > N.

1 -TT
-lJ Q,(t) dt + 1" lJ
Q,(t) dt <e. ( 1.30)

You should ponder here a bit to see what these conditions mean. The first three conditions
say that each Q, is positive. even, and normalised so that its integral is I. The last condition
says that as n becomes large the graph of Q, peaks more and more sharply at 0.

Theorem 1.3.1. Let f be a continuous function on [-rr. rr] and let Q, be a Dirac se-
quence. Then Q, * f converges to f uniformly on [-rr. rr] as n ~ oo.

Proof Let h, = Q, *f. Using ( 1.29) we have

j(

h,(x)- f(x) =
1
-rr [f(x- t)- f(x)]Q,(t) dt.

Let e be any given positive real number. The function f is uniformly continuous on
[ -rr. rr ]; i.e., there exists aD > 0 such that
e
lf<x- t)- f(x)l < 2 whenever ltl <D.

Let M = SUP-rr:::x:::rr lf(x)l. Using (1.30) we can choose N such that for alln > N,

1 -lJ
-rr
Q,(t) dt + 1" lJ
Q,(t) dt < -e.
4M
 22 FOURIER SERIES

Hence

-lJ f;r) 1/(.r- e e

(1 -rr + }lJ t)- f(x)IQ,(t)dt <2M M =
4 2.
Also.

,s 1lJ e e
1
-lJ lf<x - t ) - j(x)IQ,(t)dt < -lJ 2Q,(t)dt <
2.
Hence

lh,(.r)- j(.r)l < e for all n > N.

•
Later on. you might study the theory of distributions (or generalised functions) where
you will come across Dirac's 8- function, which is not a function in the sense we understand
it now, and which serves as an identity for the convolution operation.
In the same way. we define a Dirac family as a family of functions Qr. 0 < r < I.
where each Q,. satisl1es the conditions ( 1.27), ( 1.28) and ( 1.29) and further for each e and
8 > 0. there exists ro such that for r > ro.

1 -lJ
-;r
Q,.(t)dt + 1"lJ Q,.(t)dt <e. ( 1.31)

In this case Qr *f converges to f uniformly as r ___,. I.

Exercise 1.3.2. Show that the Poisson Kernel defined by ( 1.26) has the following proper-
tics:

(i) P(r. cp) > 0,

Cii) P(r. cp) = P(r. -cp),
(iii) P(r. cp) is a monotonically decreasing function of cp in [0. rr ],
(iv) maX-rr:::tp:::rr P(r, cp) = P(r. 0) = (I + r)/2rr( I - r),
(v) min_rr:::tp:::rr P(r. cp) = P(r. rr) = (I - r)/2rr( I + r).
(vi) f~rr P(r. cp)dcp = I.
(vii) for each e, D> 0, there exists 0 < ro < I such that, for ro < r < I. we have

-lJ
P(r. cp)dcp +
111' P(r. cp)dcp < e.
1 -TT lJ
Hint: To prove (vii) proceed as follows. Since P(r. cp) is an even function of cp you need to
prove J.;
P(r. cp)dcp < ef2. Since P(r. cp) is monotonically decreasing.

I I - r2
max P (r. cp) = - ,.
•S:::tp:::rr 'J.rr I - 2r cos 8 + r-
HEAT CONDUCTION AND FOURIER SERIES 23

Now for a fixed D > 0.

. I - ,.2
hm ., = 0.
r-+ I I - 2r cos D+ r-
Use this to choose the ro that is required.

-1T 1T

FIGURE 1
The Poisson kernel

Theorem 1.3.3. (Poisson's Theorem) Let f be a continuousfimction on T and let u(r. 0)

be defined by ( 1.25). Then as r ~ I. u(r, 0) 4> J(O) uniformly.

Proof For 0 < r < 1. let Pr(cp) = P(r. cp) denote the Poisson kernel. Then

u(r. 0) = ( Pr * /)(0).
The family Pr is a Dirac family (by the preceding exercise). Use Theorem 1.3.1 now. •

Let us sum up what we have achieved so far. We have shown that u(r. 0) defined
by ( 1.25) satisfies the conditions ( 1.2), ( 1.3) and (1.4) of the Dirichlet problem and fur-
ther (for every continuous function f) it satisfies the boundary condition (1.5) in that
limr- 1 u(r, 0) = f(O) and the convergence is unifonn in 0.
So we have found a solution to the Dirichlet problem. We will now show that the solu-
tion is unique, i.e., ifv(r,O) is another function that satisfies (1.2), (1.3) and (1.4) and if
v(r. 0) converges to f<O) unifonnly as r ~ I, then v(r. 0) = u(r, 0).

Exercise 1.3.4. Let f and g be continuous functions on T. Show that if j(n) = g(n) for
all integers n, then f = g. (Hint: Use Poisson's Theorem).
 24 FOURIER SERIES

Exercise 1.3.5. Let f be a continuous function on T. Suppose the Fourier series of f

converges uniformly. Show that its limit must be f. (Hint: Calculate the Fourier coefficients
of the limit function.)

Warning: We have not proved that the Fourier series of a continuous function on T
converges uniformly, or even pointwise, to f. In fact, this is false: there exists a continuous
function whose Fourier series diverges on an uncountable set. However, we have shown
that for each 0 < r < I. the series L~-oo ,-In I/(n )e;"e converges and as r approaches I
it converges uniformly to f.

Exercise 1.3.6. Let f be a continuous function on T and suppose the series L~-oo Ij (n) I
converges. Show that the Fourier series off converges to f uniformly on T.
This gives a sufficient condition for the convergence of Fourier series. Much stronger
results will be proved in Chapter 2.

Theorem 1.3.7. (Uniqueness of the solution to Dirichlet's problem) Let f be a contin-

uous periodic function with period 2rr. Let u (r, 0) be a function that satisfies ( 1.2), ( 1.3)
and (1.4) and let u (r, 0) con\'erge to f (0) uniformly as r ~ 1. Then u must be gi\•en
by ( 1.25).

Proof For a fixed r < l,let A 11 (r) denote the Fourier coefficients of u(r, 0), i.e.,

I
An(r) = ~ iTT u(r, O)e-1110
. dO.
_T( -TT
Differentiate with respect tor:

I I
A 11 (1') = ~ iTT u,.e- 111. 0 dO.
_T( -TT
(rA 11I (r)) I = ~I iTT (ru,.),.e- . 0 dO 111

-T! -TT

I iTT I .
= -~ -uooe-1110 dO.
-T! -TT r

since u satisfies ( 1.2). Here u,. and ue denote the derivatives of u with respect to r and 0.
Now integrate by parts and use ( 1.4) to get
')
I I n-
(/"A11(f)) = -A 11 (r).
r
I.e.,

r 2A"(
11 r) + rA I
11
(r) -n-., A 11 (r) = 0. ( 1.32)

Now this is exactly the Cauchy type differential equation which we have seen earlier
as ( 1.11 ). For n ;j= 0 its solution is

An (r) = Cnrl"l + Dnr-1"1

HEAT CONDUCTION AND FOURIER SERIES 25
where C, and D, are constants. Since

IA,(r)l <sup lu(r, {-})I

and u (r, e) is bounded on D we must have

A II (I·) -_ CII 1.1111 • /l # 0.

The C, are determined as follows:

111
C,r 1 = A,(r) = ::;--
I iTT u(r. {J)e- . 11117
lJ
dfJ.
_T( -TT
Letting r ~ I this gives

C, = ?I l;r .
[(B)e- 1110 dO = j(n), n # 0.
A

_T( -7(

Thus A, (r) = /Cn)rl"l. n =ft 0. For n = 0 the only bounded solution of ( 1.2) is

Ao(r) =constant.

Hence
7(

1 -:r u(r. 8) de is independent of r.

So for all 0 < r < I.

I
Ao(r) = Ao = -
2rr
l:r
-rr
1
u(r.B)de = -
2rr
l:r-rr
f<{J)de.

We have shown that the Fourier coefficients of u(r. 8) are j(n)rl"l. But these are also the
Fourier coefficients of ( 1.25). So the theorem follows from Exercise 1.3.4. •

Exercise 1.3.8.

(i) Prove the "mean value property" of temperature: at steady state the temperature at
the centre of a disk is the average of the temperature on the boundary.
(ii) Prove the "maximum principle'' and the "minimum principle'': at steady state the
hottest and the coldest points on a disk are at the boundary.
(iii) Can you relate these results to facts which you might have learnt in your Complex
Analysis course?

Exercise 1.3.9. Use Poisson's Theorem to prove the Weierstrass Approximation Theorem
in the following form: let /(8) be a continuous periodic function of period 2rr, then f is a
unifonn limit of .. trigonometric polynomials," i.e., finite sums of the form E;;=-Na,e;,n
Exercise 1.3.10. Let f and g be two integrable functions on [-rr, rr ]. Show that their
convolution f * g is also an integrable function. If either f or g is continuous, show that
f * g is continuous; and if either for g is C 1, show that f * g is also C 1
26 FOURIER SERIES

Exercise 1.3.11. (An extension of Theorem 1.3.1 to integrable functions). Let f be inte-
grable on [ -rr. rr] and let Q, be a Dirac sequence. Show that

(i) (Q, * f)(x) ~ f(x) iff iscontinuousatx;

(ii) if the left-hand and the right-hand limits off at x exist, denote them by f(x_) and
f<x+> respectively. and show (Q, * f)(x) ~ !rJ<x+) + f(x-)];
(iii) iff is continuous for each x in a closed interval/. then

(Q, * f)(x) ~ f(x) uniformly on I.

 . -;. ,\

.• . -. '~·:::~>+. .

Convergence of Fourier Series

We have defined the Fourier coefficients off as

/(n) = 2~ ~~ f<O)e_;,,o dO. (2.1)

These are well defined for each continuous function on T, or more generally, for each
integrable function on T The Fourier series off is the series
00

L
11=-00
/(n)i 110 • (2.2)

Associated with the series is the sequence of its partial sums

N
SN<f: 0) = L /Cn)ei
110
• (2.3)
11=-N

N = 0. I. 2..... If at a pointe ofT the sequence (2.3) converges. we say that the Fourier
series (2.2) converges at 0. It would have been nice if such convergence did take place
at every point (}. Unfortunately. this is not the case. There are continuous functions f
for which the series (2.2) diverges for uncountably many e. Now we can proceed in two
directions:
(I) Weaken the notion of convergence. or
(2) Strengthen the conditions on f.
In the first direction we will see that the Fourier series of every continuous function con-
verges in the sense of Abel summability and Cesii.ro summability, both of which are weaker
notions than pointwise convergence of the sequence (2.3 ). In the second direction we will
see that iff is not only continuous but differentiable, then the series (2.2) is convergent
at every point e to the limit f (0). More generally, this is true when f is not necessarily
differentiable but is Lipschitz or is of bounded variation.

27
28 FOURIER SERIES

2.1 Abel summability and Cesaro summability

Consider any series
00

Lx, (2.4)
II= I

with real or complex term x,, n = I. 2..... If for evel)' real number 0 < r < I the series
L~ 1 r" x, converges and if L~ 1 r" x, approaches a limit L as r ~ I. then we say that
the series (2.4) is Abel summable and its Abe/limit is L.

Exercise 2.1.1.
(i) If the series (2.4) converges in the usual sense to L, show that it is also Abel summable
to L.
(ii) Let x 11 be alternately I and -I. Then the series (2.4) does not converge but is Abel
summable and its Abel limit is I/2.
(iii) Let x, = (-I)"+ 1n. Show that the series (2.4) is Abel summable and its Abel limit
is 1/4.
(iv) In your Complex Analysis course you might have come across this notion while
studying the radius of convergence of power series. (See, for example. ''Abel's Limit
Theorem" in Complex Analysis by LV. Ahlfors.) Is there any connection between that
theorem and what we are doing now?

Poisson ·s Theorem (Theorem 1.3.3) proved in Chapter I can now be stated as:

Theorem 2.1.2. if f is a continuous function on T. then its Fourier series is Abel

summable and has Abe/limit f(O) at evel)' e.

Cesaro convergence is defined as follows. For the series (2.4) let

N
SN = L.r,, N =I, 2....
II= I

be the sequence of its partial sums. Consider the averages of these partial sums:

a,=
Sl + .\'_2 + · · · + S 11

/l

If the sequence a, converges to a limit L as ll ~ oo, we say that the series (2.4) is
swnmable to L ill the sense of Cesaro, or is Cesii.m summable to L. This is sometimes also
called (C. I) summability or swnmability by the method of the first arithmetic meall.

Exercise 2.1.3.

(i) If the series (2.4) converges in the usual sense to L, show that it is Ces~lro summable
to L.
CONVERGENCE OF FOURIER SERIES 29

(ii) If the series (2.4) is Cesaro summablc to L. show that it is Abel summable to L.
(iii) Show that the converses of statements (i) and (ii) arc false. (See the examples in
Exercise 2.1.1.)
(iv) If x, > 0. show that the series (2.4) is Ces~u·o summable if and only if it is convergent.

We will soon prove that iff is a continuous function on T. then its Fourier Series is Cesl1ro
sum mabie to the limit /(0) for every 0. In view of the relationship between Cesaro conver-
gence and Abel convergence (Exercise 2.1.3) this is a stronger result than Theorem 2.1.2.
However, it is still fruitful to usc Abel convergence
.... because of its connection with the
theory of complex analytic functions.

2.2 The Dirichlet and the Fejer kernels

For each integer n let e, (t) = ei111 • Iff is a continuous function on T. then

(f * e.,)(IJ) = ~~ f(t)e.,(O -I) dt

= e;,o lrr f(t)e-im dt
-TT

= 2rr j (n )ei"o.

Hence. we can write the partial sums (2.3) as

I N
SN (f; 0) = -?rr ""
L (f * e, )(0) = (f * DN )(0). (2.5)
- 11=-N

where

(2.6)

The expression DN(t) is called the Dirichlet kernel.

Exercise 2.2.1. Show that:

. ~ ,., sin (Nt /2) iCN+l>t/'2 for all t =ft 2krr. (2.7)
(I)L e = e •
11
=1
sin (t /2)

I
(ii) < , for all t =ft 2krr.
- I sin (t /2) I

N I sin ((N + I /2)t)

(iii) Lcosnt = -?_ + ? . ( f?)
_sm t -
. for all t =ft 2krr.
II= I

(1v. ) L e
N .
1(211-l)t _
-
sin Nt .
.
sm t
e'N' . for all t =ft krr.
II= I
30 FOURIER SERIES

(a) D, (0)

(b) D10(8)

-1T 1T

(c) DsoW>
FIGURE 1
The Dirichlet kernel for different values of n

N sin 2 Nt
(v) Z::sin ((2n- l)t) = . . for all t =ft krr.
ll=l smt

Hint: Write the sum in (2.7) as

. 1 _ eiNt
e'' . .
I - e11

then use 1 - ei' = eitf2(e-itf2 - eit/2).

Taking real parts of the two sides in (2. 7) we get (iii).
CONVERGENCE OF FOURIER SERIES 31

Proposition 2.2.2. The Dirichlet kernel has the properties

(i) DN ( -t) = DN (t),

(ii) ~~ DN(I) dt =I.

I sin(N + -.1, )t
(iii) DN(t) = - -
2rr sin t /2

Proof Properties (i) and (ii) are obvious and (iii) follows from 2.2.1 (iii).
•
The Fejer kemel F11 (t) is defined as

I 11-1
Fll(t) =-
ll
L Dk(t).
k=O
(2.8)

Exercise 2.2.3. Show that:

(i) F,, ( -t) = Fll (t).
(ii) ~~ F.,(t) dt = I.

... I sin 2 nt/2 I 1-cosnt

(m) F11 (t) = - ., =- .
2nrr sin- t 12 2nrr I - cos t
H .mt: U se Proposihon
. . -·-·-
? ? ? an d Exerc1se
. -·-·
? ? I ( v) .

(iv) F11 (t)>Oforallt.

I
(v) F11 (t) < for 0 < D <It I < rr.
nrr (I - cos D)

(vi) F11 (t) =

.
E (1 -!L!)
I
J=-(11- )
ll
eiit.

These properties of F,, can be used to conclude:

Theorem 2.2.4. The sequence F11 is a Dirac sequence.

This has a most important consequence:

Theorem 2.2.5. (Fejer's Theorem) Let f be a continuous function on T. Then the

Fourier series off is Cesaro swnmable to fat every point ofT. Further, the convergence
of the sequence

I 11-1
a~~(f; ()) =- L Sk(f; ())
ll k=O

to f(()) is unifonn on T.

Proof Note that a11 (f; ()) = (f *F 11 )(()). Use Theorem 1.3.1.
•
32 FOURIER SERIES

-1T 0
(a) F, (0)

-1T 0 1T

(b) F10(8)

-1T

(c) Fso<8>
FIGURE 2
The Fejcr kernel for different values of n
CONVERGENCE OF FOURIER SERIES 33

In (2.5) we observed that SN(f; 0) = (f * DN )(0). So if DN were a Dirac sequence we

could have concluded that the Fourier series off converges to f. However, this is not the
case. At this point we will do well to compare the three kernels P,.(O). D,(O) and F,(O) of
Poisson. Dirichlet and Fejer which we have come across. The following features should be
noted:

I. P,. (t) and F, (t) are nonnegative for all t but D, (t) is not.
2. All three are even functions, all attain a maximum at t = 0. But P,. (t) decreases mono-
tonically from 0 to rr. whereas D, (t) and F, (t) are oscillatory.
3. For D,(t) and F,(t) the oscillations become more and more rapid as n increases. How-
ever, D, (t) does not die out at ±rr, whereas F, (t) dies out at ±rr for large n. P,. (t) also
approaches 0 fort = ±rr as r ~ I.
4. The peak of all three at t = 0 goes to oo as r ~ oo.
I, or n ~

5. It is true that the integrals of all three over the interval [-rr. rr] are I. Since P,. and F,,
are nonnegative we have f~rr IP,.(t)l dt = f~rr IF,(t)l dt = I. However. we will see
that f~rr ID, (t) I dt goes to oo with n.
6. Both D, and F, are polynomial expressions in ei'. i.e., they are finite linear combina-
tions of ei"'. Such expressions are called exponential polynomials. The Poisson kernel
is not of this type.

The numbers

L, =
1"
-i(
ID,(t)l dt =~I
_rr
1" -rr
sin(n + ~ )t
. ,; dt
sm t 1-
(2.9)

are called the Lebesgue constants. We will see that as n ~ oo. L, ~ oo at the same rate
as log n.

Exercise 2.2.6. Show that

(i) sin t < t for 0 < t < rr.

T(
(ii) t < rr sin t forO<
-
t < -.
- 2
- 2

Exercise 2.2.7. Let x, = I + ~ + · ·. + ~ - logn. Show that x, converges. The limit of

x, is called the Euler constant ~nd is denoted by y. Show that 0 < y < I. Hints: Use the
fact that j(dt/t) = logt.

(i) Show that

II

~k
i"
"'"'-< -<"'-
I dt
t
11-1
~k·
I
k=2 I k=l

From this conclude that x, is monotonically decreasing and 0 < x, < I. Sox, con-
verges to a limit y.
34 FOURIER SERIES

(ii) Use the same idea to show

1 n+l I
--<log--<-.
n+l n n
Note for alln > I we can write
3 ll
logn = log2 +log-+···+ log--.
2 n- I
Use these two facts to obtain

1 - log 2 < x., < 1 - (log 2 - D.

This shows 0 < y < 1. (It is not known whether y is a rational number. An approxi-
mate value is y = 0.57722. You are likely to come across this number when you study
the ....gamma function.)

Exercise 2.2.8. The aim of this exercise is to show that L, goes to oo like a constant
multiple of log n. This can be expressed in any of the following ways:

(a) There exist constants c, and C2 such that

Ctlogn < L, < C2logn for n > 2.

(b) The sequence L, - (4jrr 2) log n is bounded; i.e ..

4
(c) L, =---:; logn + 0(1).
rr-

Here is an outline of the proof:

. 21o 1 sin(2n + I )u
(1) L, = - - - . - - - du.
rr o sm u
(ii) By Exercise 2.2.6 (i)

" Ia if-
L, > .:. sin (2n + I )u
du.
T( 0 ll

(iii) Now write

} sin(2n + l)u du = ~
~ ~:!~~~~-! sin(2n + l)u
Ia
0 ll kO
·= _1:._· -r
ln+l'"!
- - - - - du.
ll

(iv) Each of these integrals can be estimated from below by replacing u occm,-ing in the
denominator by the larger quantity

k+1 T(

2n + I 2
After this the resulting integrals arc simple to evaluate.
CONVERGENCE OF FOURIER SERIES 35

(v) This gives

4 I
>-'"""- 2n
L"- rr 2 ~ k +I
k=O
Use Exercise 2.2.7 now.
(vi) At step (ii) above use the inequality of Exercise 2.2.6 (ii) instead, to estimate L11 from
above. Make a similar change in step (iv). Get an upper bound for L11 •

Exercise 2.2.9. Estimate L11 in another way as follows:

.
(1) L 11
21TT
= -rr
sin(n + ~)u
? . ,; du.
0 -Sm 2

(ii) Split the integral in two parts; one from 0 to I In and the other from I In to rr. By
Exercise 2.2.1 (iii)

sin(n + )u! I
+ -.
2 sin I -< ll
2

So
lfn sin(n + !>u I
loo 2 sin~
du <I+-.
- 2n

(iii) For the other integral. use Exercise 2.2.6 (ii) to bound the denominator of the inte-
grand, and replace the numerator by I. This gives

TT sin (n + )u
-----:-''-- d u < -
! rr iTT du rr
- = -(log rr + log n).
i lfn 2 sin~ 2 lfn u 2

(iv) So

L 11 < logn +log rr + ~rr (1 + .;-)

_n
.

(v) Write this in the fonn

L11 < C logn for n > 2.

2.3 Pointwise convergence of Fourier series

By the Weierstrass approximation theorem every continuous function on T is a uni-
fonn limit of exponential polynomials. (See Exercise 1.3.9; this follows also from Theo-
rem 2.2.5). In other words iff is a continuous function on T. then for every E > 0 there
exists an exponential polynomial
N
PN(t) = L ani"' (2.1 0)
n=-N
36 FOURIER SERIES

such that

sup 1/(t)- PN(t)l <E. (2.11)

-TT :::£1:::£TT

This can be used to prove:

Theorem 2.3.1. (The Riemann-Lebesgue Lemma) Iff is a continuous .function on T.

then

lim f (n) = 0.
lnl-oo

Proof We want to show that given an E > 0 we can find an N such that for alllnl > N we
have 1j (n) 1 < E. Choose p N to satisfy (2.10) and (2.11 ). Note that for In I > N. fJ (n) = 0.
Hence for In I > N we have
A

f(n) = f(n)-A

p(n) = (f- -

p)(n).

But from (2.1 0) we get

- I iTT 1111
.
l<f- p)(n>l = rr -TT [f(t)- p(t)]e- dt
2

I " -
-< -E.-7r-
_rr
')
E.
•
Exercise 2.3.2. Let f be a continuous function on T. Show that

lim iTT f(t)cosnt =0. lim iTT f(t) sinnt = 0.

n-ov n-oo
-TT -TT

Exercise 2.3.3. Let f be a continuous function on T. Show that

nl!':J., L {(I) sin ( (n + Dt) dt = 0.

Hint: Use the two statements of Exercise 2.3.2 replacing .f(t) by f(t) sin~ and f(t) cos~
respectively.

Exercise 2.3...1. If f is continuously differentiable on [a, b], use integmtion by parts to

show that

lim
•\-+X
lb
ll
f(t) sin tx dt = 0.

This gives another proof of the Riemann-Lebesgue Lemma for such functions. More gen-
erally. use this method to prove this lemma for piecewise C 1 functions. [Definition: A
function f on [a. b] is called piecewise continuous if there exist a finite number of points
CONVERGENCE OF FOURIER SERIES 37

aj.a =ao <a1 <···<a, =b,suchthat.fiscontinuousoneachinterval(aj.Oj+dand

the limits f(aj+) and f(aj-) exist for all j. Such a function is called piecewise C 1• if the
derivative f' exists and is continuous on each interval (a j. a i+ I) and the limits f' (a j +)
and f' (a j-) exist for all j .]

Exercise 2.3.5. Note that the Fourier coefficients / (n) are well-defined (by the rela-
tion (2.1 )) for functions f which are integrable on T. (The term ..integrable" may be
interpreted as Riemann integrable or Lebesgue integrable depending on your knowledge).
Prove the Riemann-Lebesgue Lemma (and its corollaries) for integrable functions f. Hint:
Iff is integrable, then for every£ > 0 there exists a continuous function g such that

~~ lf(tl- g(t)i dt <e.

Notice, in particular. that piecewise C 1 functions are integrable and this will give you
another proof of the second part of Exercise 2.3.4.
These corollaries too are sometimes called the Riemann-Lebesgue Lemma.

Theorem 2.3.6. Let f be a continuous (or more generally. an integrable) fimction on T.

Let 0 < D < rr. Then for every e.
lim
(1 1rr) -.s + f(O- t)D,(t) dt = 0. (2.12)
II-OO -TT ,5

Proof Fix e. and define a function g on T as

0 ifltl <D,
g(t) = { f(O- t)
if D < It I < rr.
sin t /2
. . ') ') ?( ... )
By Propositlon -·-·- 111 ,

(1-rr
-lJ
+ frr) f(O- t)D,(t)dt =
J.s 2
I
rr
1rr-rr g(t) (sin(n + I/2)t) dt.

Now note that g is integrable and use Exercise 2.3.3 and its extension in Exercise 2.3.5 .
•
Remark 2.3. 7. In particular. choosing f to be the constant function I we get

lim
, _ 00
(1-!J + l.sfrr) D,(t) dt = 0.
-rr

This property is like Property (iv) of a Dirac sequence Q, defined in Chapter I. However
D, is not a Dirac sequence. What this property signifies is that as n becomes large the
oscillations of D, outside any D-neighbourhood of 0 cancel each other out.
38 FOURIER SERIES

Remark 2.3.8. Theorem 2.3.6 is called the Principle of Localisation. Since

S,(f: 0) = ~~ f<O- t)D,(t) dt,

(2.12) shows that the convergence properties of f at e are completely detennined by the
values of j' in a iS-neighbourhood of e, where t5 can be arbitrarily small. So the study of
convergence of S11 (f; 0) is reduced to that of the integral

i lJ
-lJ
f(O - t)D11 (f) dt

for arbitrarily small tS.

This principle can also be stated as follows: iff is zero in a neighbourhood of e. then
S11 (f; 0) converges to zero. Another statement expressing this is: iff and g are equal in
some neighbourhood of e. then the Fourier series off and g ate are either both convergent
to the same limit or are both divergent in the same way.
Now we can prove one of the several theorems that ensure convergence of the Fourier
series off when f satisfies some conditions stronger than continuity.
We say that f is Lipscltit: continuous ate if there exists a constant /vi and a t5 > 0 such
that

lf<O)- f(t)l < lv/10- tl if IB- tl < tS. (2.13)

This condition is stronger than continuity but weaker than differentiability off ate.

Theorem 2.3.9. Let f be an integrable function Oil T. Iff is Lipschit: continuous at e.

then

lim S11(f; 0) = f(O).

11-00

Pmof We want to prove

lim iTT f(O- t)D11 (t) dt = f(O).

11-00
-TT
By Proposition 2.2.2(ii) this amounts to showing

lim iTT [f(O- t)- f(O)]D 11 (t)dt = 0. (2.14)

11-00
-TT
Choose t5 and /vi to satisfy (2.13). Using Proposition 2.2.2 (iii) observe that if It 1 < tS. then
for alln,

I [f(O- t)- .f(O)]D11(t)l <.!..tv/ t

12
rr sin t /2
CONVERGENCE OF FOURIER SERIES 39

Hence for 0 < E < 8 we have

1:. [f(O- t)- /(fi)]D,(t) dt < Ce

for some constant C. Now. use Theorem 2.3.6 to get (2.14).

•
Exercise 2.3.10. Let f be piecewise C 1 on T. Show that

lim s, (f: e) = f (e) if f is continuous at e.

11-+00

and

.
Iun S, <f·, e)_ fW+>
- - -- +-f(e_)
-
,_oo 2

iff is discontinuous at e. and f(e+) and J(e_) are the right and the left limits off at e.
Hint: Write

S,(J;e)= Jo[" {(e+t)D,(t)dt+ Jo{1T /(0-t)D,(t)dt.

S,(f; e)- ~[f(O+) + f(O_)] = {" [f(O + t)- f(e+)]D,(t) dt
- lo
+ {' [f(O- t)- J(O_)]D,(t) dt.
Then use the arguments of the proof of Theorem 2.3.9 to conclude that each of the above
two integrals goes to zero as n ~ oo.
It will be worthwhile to understand how the additional condition on f in Theorem 2.3.9
has helped. In the proof of Theorem 1.3.1 and that of Theorem 2.3.9 we split the integral
into two parts. However. in the first case since Q, (t) > 0, taking absolute values did
J
not affect it. In the case of D,(t). however. ID,(t)l dt becomes unbounded. In this case
one of the integrals involved goes to zero because of the localisation principle which is a
consequence of oscillations of D,(t). The other integral goes to zero because for small t
the integrand in (2.14) is bounded independently of n. For this it is necessary that f itself
should not be wildly oscillatory.
We will look at this from another angle also. As we have mentioned earlier, there exists
a continuous function f for which the sequence

N
SN(f; e)= L /(n)e;,e
11=-N

diverges for some e. In Chapter I we saw that for each real number 0 < r < 1 the series

L rl"lj(n)eiuO
converges. We can think of this as the insertion of a "convergence factor" rl"l which con-
trols the size of the terms. It is true that j (n) ~ 0 by the Riemann-Lebesgue Lemma but
40 FOURIER SERIES

not fast enough for the Fourier series to converge. The insertion of ,-In I achieves this. What
happens in Fejer's Theorem? Notice that we can write the .. Fejer sum" aN({; 0) as

I N-l
aN(f; 0) = N L Sk(f; 0)
k=O

Theorem 2.2.5 says that aN(f; 0) ~ {(0) as N ~ oo. So, here again the inse~ion
of the ..convergence factor" I - lnlfN seems to have helped. This suggests that if f(n)
themselves go to zero fast enough, then the Fourier series off might converge without any
help from convergence factors.

Exercise 2.3.11. Let f be a C 1 function on T with derivative f'. Show that

-- A

f'(n) = inf(n). (2.15)

This is one of the most important facts of Fourier analysis whose importance you will
discover as you proceed fmther.

Exercise 2.3.12. Let f be a C 1 function on T. Show that

j(n) = 0 G). (2.16)

i.e., there exists a constant A such that

A A
lf(n)l < - . n ;/= 0. (2.17)
In I

Hint: Use (2.15) and the Riemann-Lebesgue Lemma. Prove (2.16) also when f is piece-
wise C 1
Thus. whereas the Riemann-Lebesgue Lemma ensures only that j (n) ~ 0. under the
additional hypothesis off having a continuous derivative j (n) ~ 0 at least as fast as I 1n.
In fact the smoother f is the faster is the decay of j(n):

Exercise 2.3.13. Show that f is a function of class Ck on T (i.e., f has continuous deriva-
tives up to order k). then j(n) = 0( lfnk).
Again, it should be emphasized that this relation between the smoothness off and the
size of its Fourier coefficients is an important fact of Fourier analysis.
Another important class of functions for which (2.16) holds is the functions of bounded
variation. If P is a partition of [a. b ], i.e .• a subdivision of this interval as

a =to < t1 < t2 < ·-- < t11 =b.

CONVERGENCE OF FOURIER SERIES 41

let
II

v(f, P) = L lf(t;)- f(t;_,)l. (2.18)

i=l

If this sum is less than a fixed number K for every partition P. we say that f is of bounded
variation on [a. b] and then its total variation is defined as

V (f) =sup v(f, P), (2.19)

where the supremum is taken over all partitions P.

Exercise 2.3.1-1.
(i) A function f on an interval [a. b] is said to be uniformly Lipschit: if there exists a
constant K such that

lf(s)- f(t)l < Kls- tl for all s, tin [a, b].

Show that every such function is of bounded variation on [a. b].
(ii) Suppose f is a continuous function on [a. b] and has a bounded derivative on (a, b).
Then f is uniformly Lipschitz on [a, b].
(iii) Any piecewise C 1 function on [a. b) is of bounded variation.
(iv) The function

f (t) = I o.
tcos(rr/t).
t =
O<t<l
o.

is continuous (and hence uniformly continuous) on [0, I], but not of bounded varia-
tion. (Consider the partitions

I I I
0 < - < - - < ... < - < 1.)
n n-I 2
(v) Iff is of bounded variation on [a. b ]. then there exist monotonically increasing func-
tions g. h such that f = g- h.

Lemma 2.3.15. Iff is a contimwus fimction of bounded variation on T. then

/tn) = 0 G).
Proof Integrate by parts using Riemann-Stieltjes integrals:

lf(n)l = -
I iTT .
f(t)e_,,, dt =
I iTT e-lllt
. d.f(t) V(f)
-?-.- < 2rrn . •
27r -TT -Tl Ill -TT
Exercise. Prove this when f is piecewise continuous.
42 FOURIER SERIES

For us this information is useful when we apply the following important result:

Theorem 2.3.16. Let .f be an integrable function on T such that j(n) = 0( 1/n). Then
S,(f: 8) ~ [(8) at all points 8 at which f is continuous. Iff is continuous on T. this
convergence is unifonn.

To prove this we will use a sum which is in between the two sums S,(f; 8) and a,(f; 8)
introduced earlier. For each pair of integers m, n, where 0 < m < n. define

. ) _ Sm+l (f; 8) + · ·· + S, (f: 8) (2.20)

a,,( f . 8 - .
• /l-1/l

Note that we can write

(n + I)a,+l (f; 8)- (m + l)am+l (f; 8)
am,, (f: 8) = . (2.2I)
n-m
and also
"'""' n + I - Ul ·o
a,_,(f; 8) = Sm(f; 8) + ~ f(J)e' 1 - (2.22)
A • •

. n-m
Ill< IJ 1.:::11

One way to see that this is "in between" S, and a, is to write each of these sums as
LJ=-II a(j)j(j)eiiO. For the sum S,(f; 8) the coefficients a(j) are I for -n < j < n
and 0 for ljl > n. For the sum a,(f: 8) the coefficients a(j) drop from the value I at
j = 0 to the value 0 at ljl = n and stay 0 after that. For a111 •11 (f; 8) the coefficients a(j)
are I for ljl < m and then drop to 0 at ljl = n + I and stay 0 after that. (If you sketch the
graph of a(j) you will get a rectangle, a triangle and a trapezium, respectively, in the three
cases.)
We will look at the special case ak11 ,(k+I>11 of these sums.

Lemma 2.3.17. Let f be integrable on T. Then for each fixed integer k,

ak,.(k+l>,(f; 8) ~ f(8) as n ~ oo

at each 8 where f is continuous. Iff is continuous on T this convergence is uniform on T.

Pmof From the relation (2.21) write

ak,,(k+l>,(f; 8) = (k +I+ 1/n)a(k+l>,(f; 8)- (k + lfn)ak 11 (f; 8).

Then use Fejer's Theorem (Theorem 2.2.5). Note that you will need to generalise this
theorem to be able to prove the first statement of this lemma. This is left as an exercise .

•
Lemma 2.3.18. Let f be integrable on T and let li<j)l < Afljlfor j ;j= 0. Then for all
positive integers k, m. n with kn < m < (k + I )n we have
2A
lakll,(k+l)ll(f; 8)- Sm(f; 8)1 < k
CONVERGENCE OF FOURIER SERIES 43

Pmof From the relation (2.22) we get

"
lakn.(k+l>n(f; B)- S,(f: 8)1 < lf(j>l
k,< lii.:S(k+ I )II

(k+l>n A
<2 L
j=kn+l 1
-=-

2nA 2A
<--<-
- kn - k •
PmofofTheorem2.3.16 We are given that there exists a constant A such that l.fU>I <
Afljl for i # 0. Let E: > 0 be given. Choose an integer k such that Afk < £/4. By
Lemma 2.3.17 we can find no > k such that for all n > n 0 •

Now let m > kno. Then for some n > no we will have kn < m < (k + I )n. Hence by
Lemma 2.3.18,

2A E:
lakn.lk+l>n(f; B)- S,(f; 8)1 < k < 2·
From the above two inequalities it follows that

Is, tf; B>- f <B>I < £.

•
The most important corollaries of Theorem 2.3.16 are the following theorems:

Theorem 2.3.19. (Dirichlet's Theorem) Let f be a pieceu·ise C 1 Junction T. Then 011

the Fourier series L /(n)einO converges to f(B) at every point(} where f is continuous.
This convergence is uniform on any closed inten•al that does not contain a discontinuity
off.

Theorem 2.3.20. (Jordan's Theorem) The conclusion of Dirichlet's Theorem is \'alid.

more generally, iff is any function of bounded variation on T.

Exercise 2.3.21. Recall that a function of bounded variation is continuous except possibly
at a countable set of points: and at these discontinuities the left and the right limits exist.
Show that if(} is a point of discontinuity of f. then the Fourier series L .f<n )e;,o converges
to !ff<B+) + f(B-}.
Let us make a few more remarks about the condition j(n) = O(ljn) under which we
have proved our Theorem 2.3.16. In Chapter 3 we will see examples of functions that are
piecewise C 1 (and hence, of bounded variation) for which the Fourier coefficients decay
exactly as lin. If f is a C 1 function on T. from the Riemann-Lebesgue Lemma and (2.15)
44 FOURIER SERIES

we get a stronger result than (2.16): we have

(2.23)

i.e.• lim,- 00 j (n) 1n = 0. There is another class of functions in between C 1 functions

and functions of bounded variation that is useful, especially in the theory of Lebesgue
integration. This is the class of absolutely continuous functions.
A function f on [a. b] is said to be absolutely continuous if for every£ > 0, there exists
8 > 0 such that whenever {(a;. b;)} is a finite disjoint collection of open intervals in [a, b]
with 2:~'= 1 (b;- a;) < 8, then we have 2:i'= 11f(b;)- f(a;)l < £.

Exercise 2.3.22. Prove the following statements.

(i) Every absolutely continuous function is uniformly continuous.
(ii) Every absolutely continuous function is of bounded variation.
(iii) Iff is uniformly Lipschitz, then f is absolutely continuous.
(iv) In particular iff has a bounded derivative on (a. b), then f is absolutely continuous.

The Fundamental Theorem of Calculus says that iff is absolutely continuous on [a. b ].
then it is differentiable almost everywhere, its derivative f' is integrable, and

f(t) = [ f'(s) ds + f(a) for all a < t <b.

Conversely, if g is in L 1[a, b], then the function G (t) = J~ g (s) ds is absolutely continu-
ous, and G' = g almost everywhere. (See e.g .• H.L. Royden, Real Analysis.)

Exercise 2.3.23. Show that iff is absolutely continuous, then j(n) = o(lln). In particu-
lar, this is true iff is a continuous function that is differentiable except at a finite number of
points, and the derivative f' is bounded. (This is the case that arises most often in practice.)

Exercise 2.3.2-1. In Exercise 1.3.1 0 we saw that the convolution f * g inherits the better of
the smoothness properties off and g. This idea goes further. Show that if g is of bounded
variation, or is absolutely continuous. then f * g has the same property.

We add, as a matter of record, that there exist continuous functions of bounded vari-
ation for which f # o(I In). (The standard example of a continuous. but not absolutely
continuous. function of bounded variation involves the Cantor set.)

2.4 Term by term integration and differentiation

J
If L f,,(x) = f(x), and the series converges unifom1ly, we can obtain .f by integrating
the series tem1 by term. For Fourier series we can perform tem1 by tenn integration even if
the series does not converge at all.
Let Af be any piecewise continuous (or, more generally, integrable) function on T. Let
c, = f (n ). Let
 CONVERGENCE OF FOURIER SERIES 45

F(x) =fox {(I) dt- cox. -JT <X < JT.

Then F is absolutely continuous. F(O) = 0 and F(rr) = F( -rr ). Since F is absolutely

continuous. its Fourier series

F(x) = L A,i"x
11=-oo

converges unifom1ly. The coefficients c, and A, are related by the equation c, = in A,.
From this we get

. c,
A, = -1 - for n f:. 0.
n
The condition F(O) = 0 then shows

A0 =I~-.
. "'""c,
11i=O n

This shows
x . c, . c, .
fo0 {(t) dt =cox+ 1 L-- L -e"'·r. 1 (2.24)
II i=O n II i=O n

This is exactly what we would have obtained on integrating the series

00
f(t) = L c,e;,, (2.25)
11=-':X!

term by term from 0 to x. We have proved the following.

Theorem 2.4.1. Let f be anyfimction in L 1(T) with Fourier series (2.25). Then thefimc-
tion for f (t) dt is represented by the series (2.24 ). The latter series is unifonnly conveq:ent
on T. even though thefonner may be divergent at some points.

Note that the series (2.24) is not quite a Fourier series because of the presence of the
first term. Subtracting this term we get the Fourier series for the periodic function F(x ).
How about term by term differentiation?

Exercise 2.4.2. Let f be a continuous and piecewise C 1 function on T Show that if f'
is piecewise C 1, then the Fourier series f (t) = L
j (n )ei111 can be differentiated term by
!
term. and the series so obtained converges pointwise to [f' (t +) +f' (t _) ]. (Use Dirichlet's
Theorem.)

The conditions of Exercise 2.4.2 are satisfied by functions like f (t) = It I and f (t) =
I sin tl.
46 FOURIER SERIES

2.5 Divergence of Fourier series

We have informed the reader earlier that there exists a continuous function whose Fourier
series diverges at some point. An example of such a function was constructed by Du Bois-
Reymond in 1876. This example came as a surprise because it was generally believed by
Dirichlet and, following him, by other mathematicians like Riemann and Weierstrass that
the Fourier series of a continuous function should converge at every point.
To construct this example we make use of the fact that the Lebesgue constants L 11 tend
to oo. We start with trigonometric polynomials with large Fourier sums, and combine them
to get successively nastier functions whose limit behaves more wildly than any function
whose formula could be explicitly written down. This idea is called the method of conden-
sation of singularities.

Exercise 2.5.1.

(i) Let A be any positive real number. Choose N so that the Lebesgue constant LN > A.
=
Let gN(t) sgn DN(t); i.e., gN(t) =
I if DN(t) > 0 and gN(f) =
-1 if DN(f) < 0.
Then the Fourier sum

(ii) The function 8N is a step function having a finite number of discontinuities in

[-rr, rr]. For each e > 0 we can find a continuous function g such that lg(t)l < 1 and
f::_rr lg(t) -gNU) I dt < e. (Sketch a piecewise linear function with this property.)
Use this to show that there exists a continuous function g on [ -rr, rr] with lg(t) 1 < I
and ISN(g; 0)1 >A.
(iii) Use the Weierstrass approximation theorem to show that there exists a trigonometric
polynomial p such that lp(t)l < I for all tinT and ISN(p; 0)1 > A. In other words,
for each A > 0, there is a natural number N and a trigonometric polynomial p(t) =
'\'M A( ) illl , sue h t hat
~~~=-M p n e

L
11=-M
p(n)eillt < I for all t E T,

but

L
11=-N
j)(n) >A.

Theorem 2.5.2. (Du Bois-Reymond) There exists a continuous function f on T whose

Fourier series diverges at the point 0.

Pmof Using Exercise 2.5.1 (iii) we can find, for each k = 1, 2, ... , a trigonometric poly-
nomial
CONVERGENCE OF FOURIER SERIES 47
m(k)
Pk = L pk(j)eijt (2.26)
j=-m(k)

such that

lpdt)l < 1 for all t E T (2.27)

and a positive integer n(k) such that

ll(k)
L pk(j) > i2k_ (2.28)
j=-ll(k)

We may assume, by adding zero tenns if necessary. that m(k) > n(k) and m(k) > m(k-1 ).
Now let
k
r(k) = Ll2m(j) + 1]
j=l

and

f,,(t) = L ?k1 e'r(k)t

II .
pk(t)
k=l-

(2.29)

From this it is clear that if n > k and ljl < m(k). then
A 1
f,,(r(k) + j) = 2k pk(j). (2.30)

and
A

f,,(j) = 0 for all j < 0. (2.31)

If n' > n + 1. then

II'

I J+II' (t) - J+II (f )I = '"""' ' eir!k>tPk(t)

~ ,..1k
k=ll+l -

II ' 1
< L ')k IPkU)I
k=ll+l -
I
II 1
= L ')k ·
k=ll+l -

So by the Weierstrass M-test the sequence f,, converges unifonnly on T to a continuous

function f. The properties (2.30) and (2.31) are carried over to f. Using these two proper-
48 FOURIER SERIES

ties we see that

r(k)+n(k) l"lkl-n(k)

L L
A

iu>- f(j)
j=O j=O

ll(k)
L f<r(k) + j) - f<r(k) - n(k))
j=-ll(k)

1 ll(k)
= k
2
L fh(j)- fh(n(k))
j=-ll(k)

1 ll(k)
> k
2
L f1k(j) -lfJdn(k))l
j=-ll(k)

I .,k
> -(2- - 1)
- 2k

using (2.27) and (2.28). Ask~ oo this expression goes to oo. Thus the sequence SN(.f; 0)
cannotconverge. •

Note that our proof shows that lim SN(.f; 0) = oo.

The method of condensation of singularities has been found to be very useful in other
contexts. Some of its essence is captured in one of the basic theorems of functional analysis
called the Uniform Boundedness Principle or the Banach-Steinhaus Theorem (discovered
by Lebesgue in 1908 in connection with his work on Fourier series; and made into a general
abstract result by Banach and Steinhaus). We state the principle and then show how it can
be used to give another proof of Theorem 2.5.2.

The Uniform Roundedness Principle. Let X and Y be Banach spaces and let A, be a
sequence of bounded linear operators fmm X to Y. {f the sequence II A,x II is bounded for
each x in X, then the sequence II A,ll is also bounded.

Let X be the Banach space C(T) consisting of continuous functions on T with the norm
of such a function defined as

llfll =sup lf(t)l. (2.32)

lET

Now define a sequence of linear functionals on X as

A,(.f) = S,(.f; 0). (2.33)

CONVERGENCE OF FOURIER SERIES 49

Note that

IAn<f)j = ~~ /(t)D.,(t)dt

< IIIII ~~ ID.,Ctll dt

= L, llfll. (2.34)

where L, is the Lebesgue constant. In particular this shows that

II A, II < L, for each n. (2.35)

We will show that

II A, II = L, for each n. (2.36)

Fix n. Let g(t) = sgn D,(t). As seen in Exercise 2.5.1. we can choose a sequence¢,
in C(T) such that ll/>m(t)l < I and lim,- 00 ¢,(1) = g(t) for every t. Hence, by the
Dominated Convergence Theorem,

lim A,(¢111 ) = lim

m-oo m-oo
lrr
-TT
¢,(t)D,(t)dt

= ~~ g(t)Dn(t)dt
= ~~ IDn(t)idt.
Since 11¢, II = I. this proves (2.36). So II A, II is not a bounded sequence. Hence by the
Uniform Bounded ness Principle there exists an f in the space X = C( T) such that the
sequence

IA,(/)1 = IS,(f; 0)1

is not bounded. Therefore, the Fourier series off at 0 does not converge.
Let us make a few remarks here. The point 0 was chosen just for convenience. The same
argument shows that for each pointe on T we can find a function f such that IS,(/: 8)1
is not bounded. The Uniform Boundedness Principle is (usually) proved using the Baire
Category Theorem. Using these methods one can see that the set of all functions whose
Fourier series converge at 0 is a set of first category in the space C(T). In this sense
continuous functions whose Fourier series converge everywhere form a meagre set. Using
a little more delicate analysis one can show the existence of a continuous function whose
Fourier series diverges except on a set of points of the first category in T. and the existence
of a continuous function whose Fourier series diverges on an uncountable set. (See W.
Rudin, Real and Complex Analysis, Chapter 5).
One may now wonder whether it is possible to have a continuous function whose Fourier
series diverges eve1ywhere.
50 FOURIER SERIES

After Du Bois-Reymond·s result, this is what mathematicians tried to prove for sev-
eral years. In 1926. A.N. Kolmogorov constructed a Lebesgue integrable function on T.
i.e .. a function in the class L 1(T). such that IS,(f; 0)1 diverges for all e. Though Kol-
mogorov's function was not Riemann integrable, this encouraged people in their search for
a continuous function whose Fourier series might diverge everywhere. However. in 1966,
L Carleson proved that iff is square integrable, i.e., in the space L 2 (T), then the Fourier
series off converges to f almost everywhere on T. In particular iff is continuous. then
its Fourier series converges almost everywhere.
Another natural question may be raised at this stage. Given a set E of measure zero in
T, can one find a continuous function f on T such that IS11 (f; 0)1 is divergent for all e
in £?The answer is yes! This was shown by J.P. Kahane and Y. Katznelson soon after
Carleson 's result was proved.
So Fourier was wrong in believing that his series for a continuous function will converge
at every point. But he has been proved to be almost right.
Odds and Ends

3.1 Sine and cosine series

A function f is called odd if f(x) = - j(-x) and even if J(x) = J(-x) for all x. An
arbitrary function f can be decomposed as f = /even+ /odd. where /even(X) = ~[J(x) +
J(-x)] and /odd(x) = ![J(x)- J(-x)]. The functions !even and /odd are called the e1•en
part and the odd part of f, respectively. Notice that if f (x) = eix, then /even (x) = cos x
and /odct(X) = i sinx.

A A

Exercise 3.1.1. Show that iff is an even function on T, then j(n) = J(-n), and iff is
odd, then j (n) = - j (-n). In other words the Fourier coefficients are also then even and
odd functions on the integers Z.

So if f is an even function on T, then

N
SN(f; 0) = L j(n)ei118
11=-N
N
= j(O) + 2 L j(n) cos nO.
11=1

Also note that if f is even, then

j(n) = -1-lrr f(t)(cosllt- i sinnt)dt

2rr -rr
= -1 !orr f (t) cos ll t d t.
T( 0

51
52 FOURIER SERIES

Putting a, = 2j(n ), we can write the Fourier series of an even function f as

00

-::;
ao + "'""
~a, cos nO, (3.1)
- 11=1

where

a,=-I iTT f(t)cosntdt. (3.2)

T( -TT
This is called the Fourier cosine series.
In the same way. iff is an odd function on T, then we can write its Fourier sine series
as
00

L b, sin nO (3.3)
II= I

where.

b, =-I iTT f(t)sinntdt. (3.4)

T( -TT
The Fourier series of any function on T can be expressed as
00
ao "'"" cosne + b, smnO),
+~(a, . (3.5)
2 II= I

where a, and b, are as given above. In several books the series (3.5) is called the Fourier
series off and is taken as the starting point. Let us say that (3.5) is the Fourier series in
its trigonometric form and
00

L c,ei"e (3.6)
11=-00

is the Fourier series in its exponential form.

Exercise 3.1.2. If (3.5) and (3.6) are the Fourier series of the function f, show that

a, = c, + c_,. b, = i(c,- c_,),

and for n > 0,

I I
2(a + ib,).
c, = (a,- ib,), c_, =
2 11

Exercise 3.1.3. The series (3.5) can be integrated term by tenn to give
00 00
x
=-;:;- + L- + L
l lo-•,. lJ11 (
-n sinnx--l cosnx ) .
a, ')II

lo 0
f(O)dO
- 11=1 n 11=1 ll
(3.7)
 ODDS AND ENDS 53

Iff is sufficiently smooth, the series (3.5) can be differentiated term by term, and then
00

['{e)= "L,<nb, cosn8 -na, sinn8). (3.8)

II= I

(See Section 2.4)

Exercise 3.1..1. We have seen above that iff is a continuous even function on T. then
given an e > 0 there exists a finite sum g(8) = E:~=O a, cos n8 such that

sup 1/(8)- g(fJ)I < e. (3.9)

OeT

This can be used to prove the Weierstruss appmximation theorem as follows:

(i) Show, by induction, that there exists a polynomial T,, of degree n with real coefficients
such that

cosn8 = T,,(cose). (3.10)

T,, is called the Tchebychev polynomial of degree n.

(ii) Let q; be any continuous function on [0, I]. Define a function f on T by JW)
q;(l cos 81). Then f is an even continuous function. Use (3.9) and (3.10) to show that

N
sup lq;(t)- L,a,T,,(t)l <e.
O:::t:::l 11=0

This shows that every continuous function on [0. I] is a uniform limit of polynomials.
(iii) By a change of variables show that every continuous function on an interval [a, b] is a
uniform limit of polynomials. This is called the Weierstrass approximation theorem.
(iv) Show that this result and the one we proved in Exercise 1.3.9 can be derived from
each other.

Exercise 3.1.5. Use the Weierstrass approximation theorem to prove that if f and g are
continuous functions on [a. b] such that

J.b x" f(x)dx = J.b x"g(x)dx

for alln > 0, then f = g. (This is called the Hausd01f! moment theorem and is very useful
in probability theory).

3.2 Functions with arbitrary periods

So far. we have considered periodic functions with period 2rr, identified them with func-
tions on T and obtained their Fourier series. A very minor modification is required to deal
with functions of period 2L, where L is any real number. Iff is periodic with period 2L,
54 FOURIER SERIES

then the function g(x) = j(xLjrr) is periodic. with period 2rr. So. we can first derive the
Fourier series for g and then translate everything back to f by a change of variables. All
the results proved earlier remain valid with appropriate modifications.

Exercise 3.2.1. Let L be any positive number and let f be an integrable function on
[-L. Ll with f(L) = J(-L). Then the Fourier series off can be written as

L A,ei":mfL,
where

A, = - I 1L j(t)e- 1111
A . rr fL dt.
(3.11)
2L -L

In the same way. iff is even. then its cosine series can be written as
00
I nxrr
-Ao
"J
+"'"'A,
~
cos--,
L
- 11=1

where

2 {L ntrr
A, = L lo .f(t) cosT dt; (3.12)

and iff is odd. then its sine series can be written as

00
• nxrr
LB,smL.
II= I

where

21L . ntrr
B, = - [(t) sm- dt. (3.13)
L 0 L
Now we come to an interesting bit: how to apply these ideas to any function on a bounded
interval. Let J be any continuous (or piecewise C 1) function defined on the interval (0. L].
Extend f to [-L. L] as follows. Define

{(0) = J(O+>·

f(-x) = f(x) for 0 < x < L.

This gives a continuous (or piecewise C 1) function on [-L. L]. Now extend f to all of IR
by putting

f<x + 2L) = f<x> for all x.

This is an even continuous (or piecewise C 1) function defined on all of IR having period 2L.
So we can expand f into a cosine series. Notice that our extended function is continuous
ODDS AND ENDS 55

at x = 0 and x = L. We could also have extended f to an odd function as follow~. Define

j(O) = 0.
{( -x) = - j(x). 0 <X< L.

This extends f to [- L, L ]: and now extend f to IR by putting

f(x + 2L) = f(x) for all x.

Now we get an odd periodic function. Notice that now f(O) = 0 and f is discontinuous at
L with f<L-> = - J<L+>· So we can expand f into a sine series also.
This apparently anomalous situation-the possibility of expanding any function as a
sum of periodic functions and further the possibility of expanding the same function as both
a sum of even functions and a sum of odd functions-was a part of the raging controversy
at the beginning of the subject. See Chapter 0.

Examples in the next section will clarify further how this apparent anomaly is neatly
resolved by our analysis.

3.3 Some simple examples

Example 3.3.1. Let f be the "pulse function'' defined as

j(x) =
l -1
I
if- Tl <X < 0.
ifO <X < T(,

and extended periodically to all of I!t This is an odd piecewise C 1 function with jumps at
0, ±rr, ±2rr, .... The average value of f at these points is 0. The Fourier series of f is a
sine series like (3.3) with coefficients given by

b, = .:..') !oTT sin nt dt.

T( 0

So

4/ /lJT if n is odd,
b.,= { if n is even.
0

Hence we ha vc
4 (sin x
j(x) = - - + sin 3x + sin 5x +... ) . (3.14)
rr I 3 5

At x = 0. ±rr. ±2rr, ... , the series adds up to zero which is the average value off at these
points. At other points it converges to j(x). Figure 1 shows the pulse function. Its Fourier
sums for I, 2, I 0 and I00 tenns are shown in the next figure.
 -1T 1T

FIGURE 1
The pulse function

(a) s, (f; 0)

-1T
1T

(c) SIO(f; 0)

-1T
1T

(d) SlOo(J; 0)
FIGURE 2
Fourier sums of the pulse function for different values of N
56
 ODDS AND ENDS 57

Example 3.3.2. Now consider the pulse function with a different width. Let

-1 for - I < x < 0.

f(x) =
{ 1 for 0 < x < I

and extend it to IR as a function with period 2. Show that the Fourier series off is

4 (sin rr x
f (X)=- + sin 3rr x + sin 5rr x + ....
)
(3.15)
rr I 3 5

Note that the Fourier coefficients of f in Examples 3.3.1 and 3.3.2 are the same, but the
wavelengths of the sine waves have been adjusted. (The amplitude of the wave is the same
as before, the wavelength is shorter.)

Example 3.3.3. The function

0 for - I < x < 0.

f(x) = {
for 0 < x < I

is called the Hem•iside fimction. Extend it to lR as a function with period 2. and show its
Fourier series is

I 2 (sinrrx sin3rrx sin5rrx )

-+-
2 rr I
+ 3
+ 5
+··· . (3.16)

Obtain this directly from the preceding example.

Example 3.3..1. Let

rr') +x for - rr -< x -< 0'

f(x) = ;
{ - - x for 0 -< x -< rr
2

and extend f to lR as a periodic function with period 2rr. You get a ..triangular wm•e" with
amplitude rr /2 and period 2rr. Show that f has the expansion

j(x) = -4 (COS
T(
-.,-
1-
X
+ COS.,3.r +... ) .
3-
(3.17)

Example 3.3.5. Let f be defined on [0, I) as

f(x) = x. O<x<l.

Extend f to IR as an even function by the prescription in Section 3.2. This gives a saw-tooth
cun•e shown in Figure 3. This is an even continuous function with the cosine series

f(x) = -2I - --:;

4 ( cos rr x
rr- 1-
., + cos 3rr
., x +... ) .
3-
(3.18)

If we extend f to an odd function by our prescription we get the function shown in Figure 4.
58 FOURIER SERIES

FIGURE 3
The even saw-tooth curve

FIGURE 4
The odd saw-tooth curve

This has discontinuities at all odd integers, where the average value is zero. The Fourier
series is now a sine series
2 (sin rr x sin 2rr x
f(x) = -
T( 1
-
2
+ sin 33rr x - · ·· ) . (3.19)

We can extend f beyond its original domain in one more way. We put f(x) = 1 + x for
-1 < x < 0, and then extend the function to all of IR as a periodic function with period 2.
This gives the function (neither even nor odd), see Figure 5(a).

//1// 0
(a)

(b)
FIGURE 5
Two more extensions of the saw-tooth function
ODDS AND ENDS 59

We can obtain the Fourier series for this function from (3.19) in three steps. First con-
sider the odd function g(x) = x. -~ < x < ~.and extend it to IR as a periodic function
with period I. The amplitude and th~ wavelength of g are half those of the function f in
(3.19). So

I (sin2rrx sin 4rrx sin 6rrx __ . ·) .

g(x) = -
rr I 2 + 3

The function f whose Fourier series we want can be expressed as f(x) = g(x- ~) + ~-
So its Fourier series is - -

f (.\·)-~-_!_(sin2rrx
- 2 JT 1
+ sin4rrx
2
- sin6rrx
3
- ... ) .
(3.20)

A fourth way of extending the original function of this example is as follows. Let

:r for - I -< x -< 1'

j(.r.) = 2- X for I < x ~ 2.
{
-2-x for - 2 < x < -I.

Then extend f to all ofiR as a periodic function with period 4. This gives an odd triangular
wave, see Figure 5(b ). A calculation shows that f has the Fourier series

8 (sinrrx/2 sin3rrx/2 sin5rrx/2

f (x) = ---:;
rr-
., -
1-
., + ., - . ..) .
3-
(3.21)
5-

We have thus four different periodic functions that coincide with f(x), for 0 < x < I.
Their Fourier series are given by the formulas (3.18)-(3.21 ). The first function is even, the
second and the fourth are odd. the third is neither even nor odd. Two of the four functions
are continuous. The corresponding series (3.18) and (3.21) are uniformly convergent. The
series (3.19) and (3.20) converge uniformly on closed subintervals of (-I, I) and (0, 1)
respectively. At the points x = n, where n is an integer. the series (3.19) converges to the
value 0. and the series (3.20) to the value I /2.

Example 3.3.6. Let

- COSX ~f - JT <X < 0.

{(x) = 0 If X= 0.
{
COS X if 0 < X < JT

and extend f to ffi. as a periodic function with period 2rr. This is an odd function with
discontinuities at x = nrr. Show that the Fourier series off is given by

~ ~ n sin 2nx
(3.22)
rr ~ 4n 2 - 1 ·
II= I
60 FOURIER SERIES

If we differentiate this series term by term, we obtain

16 co n 2 cos 2nx
-~ ., .
rr L 4n-- 1
11=1

This series is divergent. (See Exercise 2.4.2).

Example 3.3.7. Let j(x) = x 2 , -rr < x < rr, and extend this to IR as a periodic function
with period 2rr. The Fourier cosine coefficients of this even function can be calculated by
integrating by parts twice. We get

X
2
=-+4
L (-1) cos.,nx
rr-
., co
11 for - rr < x < rr. (3.23)
3 n-
11=1

This series can be differentiated term by term (see Exercise 2.4.2) to get
co -
~ ,+ 1 smnx
x=2L(-1) for - rr < x < rr. (3.24)
ll
II= I

This is the series we have obtained in (3.19). In the reverse direction, we can integrate term
by term the series obtained for x in Example 3.3.5 and obtain different series for x 2• Thus,
the series (3.20) leads to

x2
-
2
=
.
T( _\ -
2 Leo 1
~
n-
+ 2 Leo cosnx
.,
n-
for 0 < x < 2rr. (3.25)
11=1 11=1

Integrating the series (3.18) we get

co .
..,,.2 -- T(Y--
.•
8
rr
L -n3- SlnllX for 0 < x < rr. (3.26)
II= I

These are not Fourier series. If we substitute in them the Fourier series for x (in the appro-
priate domains) given by (3.20) we obtain from (3.25)
., co co .
•2 _ 4rr- cosnx ~ ~ smn.r
.l - -
3
+ 4L ., - 4rr L
n- n
. for 0 < x < 2rr, (3.27)
11=1 11=1

using the fact L 1/n 2 = rr 2/6 (see (3.39)), and from (3.26)

x = rr
2
- - rr
2
2
L sm. 2nx - rr-8 L smnx
~ co .

'
for 0 < x < rr. (3.28)
11 3
11=1 11=1

Yet another series representation for x 2 is given in Exercise 3.5.2.

Example 3.3.8. Show that the Fourier series for the function 1sin x 1 is

.· _
I sm .l 1 = -2 - -4 Leo cos2nx
., . (3.29)
rr rr II= I
4n- - 1
ODDS AND ENDS 61

This representation is valid for all x E JR. Show that the Fourier series for the function

j(x) = l o_
smx
for - rr < x =:: 0.
forO< x < rr
IS

I
-+----
. _, - L cos'"J1
S ·In ·
.,
, x -l.\·
. (3.30)
rr 2 rr
II= I
..J.n- - I

3.4 Infinite products

Let {a"} be a sequence of real numbers. How do we define the infinite product na11 ? We
can imitate the definition of an infinite sum. For each N = I. 2.... let PN = a1a2 ···aN
be the sequence of partial products of {a11 }. If this sequence converges to a limit p and
p f:. 0. we say that the product na11 converges, and write

n!:\.)

II= I
all= p.

Note that if any a11 = 0. then the product is zero. This case is uninteresting. Further, if
p N converges, then a11 too has to converge to I. So. after ignoring the first few factors. we
may assume that a11 > 0 for all n. In this case log a11 is meaningful. and we can convert
questions about products to questions about sums.

Exercise 3..1.1. Let a11 > 0 for n = I, 2, .... Show that the product na
11 converges if and
only if the series L log a11 converges.

Exercise 3.4.2.
(i) Show that if lxl < 1/2. then
I 3
:;lxl < llog(l + x)l < :;lxl.
- -
Hint: For lx I < I, we have
x2 x3 x4
log (I +X) = X - + J - "J + · · ·
2
(ii) If a11 > -I for alln, then the series L I log( 1 + a") I converges if and only if L Iani
converges.
Let {a11 } be a sequence of positive real numbers. If the series L log a" converges
absolutely. we say that the product na11 converges absolmely.

Exercise 3.4.3. Let a11 > 0 for alln. The product na11 converges absolutely if and only if
the series :L<a11 - I) converges absolutely.
Uniform convergence of products like nJ,,(x) can be defined in the same way as for
series.
62 FOURIER SERIES

We will derive now some important product representations for trigonometric functions.
Consider the function

j(x) = cost.\'. -T( < t < T(, t not an integer.

This is an even function. Its Fourier cosine series can be determined easily. We have

2 sin trr
ao=-
T( t

( -1)" 2t sin trr

a,= T(
., ., .
t- - n-

Thus we have

2t sin trr ( I cosx

costx =
T(
.... ., - 2
Lt- t - 1-
., + t"-cos2x
., - ,...,
L-
- · · ·)
.

Choosing x = rr, we obtain

This can be rewritten as

I 2t co I
cotrrr- - = - "\""" ., , (3.31)
rr t rr L t 2 - n-
11=1

or as

rr cot rr t = -1+ L (--

I + -1) , (3.32)
t II i=O t - n /l

where the last summation is over all nonzero integers.

This is a very important formula. It is called the resolution of the cotangent into partial
fractions.
If 0 < t < a < 1, the series in (3.31) is dominated by the convergent series

I
.,
"\""" .,.
L-n . . - a-

Hence the series is uniformly convergent and can be integrated term by tenn in this domain.
Performing this integration, we see that for 0 < x < I. we have

sm rr x "\""" x-.,)
Iog . co (
= L Iog 1 - ----;; .
T!X
Il=
I n-
ODDS AND ENDS 63

Taking exponentials of both sides we get the marvelous product formula

sinrrx
T!X = n00
(
I-
x2)
n2 . (3.33)
11=1

Exercise 3.4.4. Use the relation cos rr x = sin 2rr x 1(2 sin rr x) to obtain from this the prod-
uct formula

cos T( x = n00

11=1
( 1_
(2n - I)-
4")
x-
., . (3.34)

Exercise 3.4.5. Show that

rr-
., 'XI

.
sm-rrx
., 11~oo (x - n )2.
(3.35)

Exercise 3.4.6. We have obtained these infinite series and product expansions for 0 <
x < 1. Show that they are meaningful and true for all real x. (They are valid for all
complex numbers. and are most often included in Complex Analysis courses.)

Exercise 3.4.7. Show that

I I ~
-.-=-+L.,..(-1) 11 [ I + I ] .
smx x 11 =1 x- nrr x + nrr

3.5 1r and infinite series

The number rr has allured, fascinated and puzzled mathematicians for over two millenia.
Approximations torr and formulas for it have been among the prize discoveries of famous
mathematicians. Fourier series provide an inexhaustible source for such formulas. Some of
them are given below. You can certainly discover more.
Choosing x = rr/2 in (3.14) we get
rr 1 1 I
4 =l-3+5-7+··· (3.36)

This seems to be the earliest example of a series involving reciprocals of integers and rr. It
was discovered by Nilakantha in the fifteenth century via the series
x3 x5
arctan x = x - J + 5 - ·· · , (3.37)

valid for lxl < 1. This series was known to the fourteenth century mathematician Madhava.
These series are now generally known as Leibniz-Gregory series.
Choosing x = 0 in (3.17) we get

rr 2 1 I
8 = 1 + 32 + 52 + .... (3.38)
64 FOURIER SERIES

From this we can derive Euler's wonderful and famous formula discovered in 1734.
.,
rr- 1 1
-=1+-+-+···.
6 22 32
(3.39)

Note that the series on the right is convergent and its sum is bounded by 2. This was noticed
by the Bernoulli brothers (Jakob and Johann) who asked for the exact value of the sum.
Rearranging its terms we can write

001 00 1 001
~--::;=~ .,+~-.,
L n- L (2n- 1)- L (2n)~
11=1 II= I II= I
00 I 1 00
1
= ~ +-4 ~
., ?·
L (2n - 1)- L n-
11=1 11=1

Using this we get (3.39) from (3.38). (Euler's proof was different and is discussed later.)
Choosing x = 1/4 we get from the series (3.19).

rr 1 I 1 I I
-=1+-----+-+--···. (3.40)
2J2 3 5 7 9 II

a series in which two negative signs alternate with two positive signs. Choosing x = 1/2
in (3.21) we get

rr 2 1 I 1 1
-=1----+-+--···
8J2 32 52 72 92
(3.41)

Choosing x = rr /2 in (3.26) we get

rr 3 1 I 1
-3 =l----:;-+---:;---3+··· (3.42)
2 3-' 5-' 7-
Choosing x = rr /2 in (3.29) we get

(3.43)

Exercise 3.5.1. Show that

(3.44)

Exercise 3.5.2. Use (3.25). (3.39) and a rescaled version of (3.20) to show

.,
x-=3rr-+
4 ., 4~
L
(cosnx rr sinnx)
., - - - - . (3.45)
11=1 n- 11

for 0 < x < 2rr.

ODDS AND ENDS 65

Exercise 3.5.3. From the product formula (3.33) obtain the Wallisfomwla

T( 2244
- = ---- ...
2 1335

Exercise 3.5.4. Show that

1 logx
1
rr2
--dr---
0 I -x - - 6-

3.6 Bernoulli numbers

This is an important sequence of rational numbers discovered by Jakob Bernoulli. These
numbers arise in several problems in analysis.
Let B, (x) be the sequence of functions on the interval [0.1] defined inductively by the
conditions

(i) Bo(x) = 1.
(ii) B;,(x) = B,_, (x) for n = 1. 2 .... .
(iii) Jd B, (x) dx = 0 for n = I. 2.... .
These conditions determine uniquely a sequence of polynomials called Bemoulli polyno-
mials. The first four are given by

Bo(x) = I.
B, (X) =X- 1/2.
B2(x) =x 2 j2-x/2+ 1/12.
B3(x) =x 3 j6-x 2 /4+xfl2.
2
B4 (x) = x 4 /24- x 3I 12 + x /24- 1/720.

Exercise 3.6.1. Show that

(i) B,(x) is a polynomial of degree n.

(ii) B,(x) = ( -1)" Bn (I - x) for alln > 0.
(iii) B, (0) = B, (I) for all n > 0 except n = 1.
(Note 81 (0) = -B, (1) = -1/2.)
The Bemoulli numbers are the sequence B, defined as

B, = n!B,(O). (3.46)

We should warn the reader here that different books use different conventions. Some do
not put the factor n! in (3.46) while others attach this factor to B,(x).
66 FOURIER SERIES

Exercise 3.6.2. Show that

B,(x) = -
1
n!
L II

k=O
(
ll

k
) Bkx
11-k (3.47)

Use Exercise 3.6.1 to show that Bo = I. and for n > 1.

B,=--L
I n-l (n + 1) Bk. (3.48)
n + I k=O k

The last relation can be w1itten out as a sequence of linear equations:

1 + 2B, = 0.

+ 382 = 0.
1 + 3B,

I + 48, + 682 + 483 = 0.

1 + 58, + 1082 + 1083 + 584 = 0.

From here we see that

B 1 = -1/2. B2 = 1/6. B?. = 0, 84 = -1/30. Bs = 0,

B6 = 1/42, B1 = 0. Bs = -1/30, ....

Exercise 3.6.3. Show that

e 1
t
- 1
=I: B, 1,,
n=O n!
(3.49)

(Hint: W1ite down the power se1ies for e' - 1. multiply the two series, and then equate
coefficients.) From this we see that

t t IJC B
- - + - = I + "'""'
e1 - I 2 ~ n!
___!!.. r". (3.50)
11=2

The left-hand side can be rewritten as r(e 1 + l)/2(e'- 1). This is an even function of r.
Hence we must have

B2n+l = 0 for n ~ I. (3.51)

What has all this to do with Fourier series? We have seen that the functions B, (x), n > 2
are periodic on IR with period 1. The function 8 1(.t) has the Fourier series

I ~sin 2rrnx
B ,(x)=--~ . 0 <X < I. (3.52)
JT "=' ll
 ODDS AND ENDS 67

(See (3.20)). Repeated integration shows that form > I,

") N ')
B2m (x) = ( -l)m-1 -., "'""'cos ~rrnx • (3.53)
(2rr )-111 ~ n-111
11=1

B2m+ I (.\:·) -_ (-I )m-1

(2rr)-"'+l
L
")
-
.,
sm
N • ")
_rrnx
---:.,=---~
n-m+l
(3.54)
11=1

for 0 ~ x < I. (This shows again that B2m+l = 0 for all m > 1.)
The Riemann ::.era function is the function

oc I
{(s) = "'""'---;.
~/l"
(3.55)
11=1

a meromorphic function defined on the complex plane. We will be concerned here only
with its values at positive integers. We have seen formulas for {(2) and {(4) in (3.39)
and (3.44 ). More generally we have the following.

Theorem 3.6.4. For e\'el')' positi\•e imeger m the \'Cdue of {(2m) is a rational multiple of
rr 2"' gi\•en by the formula

(-I (2rr) 2111 B.,

r (? ) _ _ _)"'-I
__ ___ -_m
~ -Ill - (3.56)
2(2m )!

Exercise 3.6.5. Show that

(3.57)

The result of Theorem 3.6.4 was proved by Euler. The corresponding question about
{(2m+ I) was left open and has turned out to be very difficult. Not much is known about the
values of the zeta function at odd integers. In 1978 R. Apery showed that {(3) is irrational.
Very recently (2000) T. Rivoal has shown that among the values {{3), {(5), ... , {(2n + 1)
at least log(n)/3 must be irrational, and that among the nine values {(5), {(7). ... , {(21)
at least one is irrational. W. Zudilin has improved this to show that at least one among the
four numbers {(5), {(7), {(9) and {(II) is irrational. These methods do not show which
of them is irrational.

3.7 sinxjx
The function sin x fx and the related integral Si (x) defined as

x sin t
= -dt
Si(x)
lo0 t

appear in several problems. The function sinxjx is even while Si(x), called the sine inte-
gral, is odd. Their graphs are shown in Figure 6.
68 FOURIER SERIES

:::::;;;;>"'""'=::::::::---.~:::::;;::;;-=........:=·.---==-
-27T -7T 1T 21T

( a) sin(x)
.l

-27T

(b) Si(x)

FIGURE 6

Exercise 3.7.1. Show that the integral

smr
dt
t

is divergent. (See the discussion in Exercise 2.2.8.)

Exercise 3.7.2. Show that the integral

00
sin r InA --dr
sin r
1 o
--dr
t
= lim
A-co o t
(3.58)

is convergent. (Split the integral into two parts

The first one is clearly finite. Integrate the second by parts. Alternately. represent the inte-
gral as an infinite sum with terms
(11+1)7T

1 111T
sinrjrdr.

These tem1s alternate in sign and decrease in absolute value.) Thus we have an example
of a function for which the Riemann integral (in the .. improper.. sense (3.58)) is finite but
which is not Lebesgue integrable over JR.
The main result of this section is the fonnula
00
sin r rr
1 0
-dt=-.
t 2
(3.59)

There are several ways to prove this. We give two proofs related to our discussion of Fourier
series. In our discussion of the Dirichlet kernel (Proposition 2.2.2) we have seen that
"sin ((n + 1/2)1)
Ino .
sm (t /2)
dt = T( (3.60)
ODDS AND ENDS 69

for every positive integer n. Let

I
----- 0< 1 :5 rr,
g(t) = { 1/2 sin (1/2>'
0 1 = 0.
This is a continuous function (use the L'Hopital rule). Hence by the Riemann-Lebesgue
Lemma (see Exercises 2.3.4 and 2.3.5)

lim [" g(1) sin ((n + 1/2)1) d1 = 0. (3.61)

11-oo lo
From (3.60) and (3.61) it follows that

. [," sin ((n

IliD
+ 1/2)1) d1 = -.
rr
(3.62)
11-00 0 1 2

Changing variables by putting x = (n + 1/2)1, we get

. [,(11+1/2)rr sinx rr
hm
11-"'00 0
--dx
X
= -.
2
(3.63)

This is the formula (3.59).

Exercise 3. 7.3. The integral I in (3.59) can be written as

sin 1
I= Loo 1(11+ I )rr /2

11=0 111T /2
-d1.
1

When n =2m, put 1 = mrr + x, to get

(1l+l>rrf 2 sin1 [,"'2 smx
1 11TT/2
-d1=(-l)"'
1 0 mT! +X
dx.

When n = 2m - I, put 1 = mrr - x to get

11 +l)rr/ 2 sin1 [,"'2 smx
1
<
- d 1 = (-1)"'- 1 dx.
11TT/2 1 0 mT! -X

Hence
2 00 2
I= rr 1 sin-
- x
dx +L [," 1
(-1)"'
[ I + I ] sinxdx.
[,
0
x Ill= 1
o x - mrr x + mrr

The integrands in the last series are bounded by the terms of a convergent series. Hence

00
I= [, rr /2 sinx { -1 + L(-1)"' [ x - Imrr + x +1mrr ] } dx.
0 x m= 1

The sum in the braces is 11 sin x by Exercise 3.4. 7. This gives another proof of (3.59).
70 FOURIER SERIES

Exercise 3.7...1. Use the Taylor expansion

x3 x5
Si(r) = r - - -
. - 3. 3!
+- - · ··
5. 5!
to calculate Si(11) using a pocket calculator. You will see that Si(11) ~ 1.852.

Exercise 3.7.5. Show that for every real number A,

sgnA =-
21
11 0
00
sin At
--dt.
t

This is an example of an imegral represemarion of a function-the function sgn A is ex-

pressed as an integral.

Exercise 3.7.6. Show that

00
sin t cost
1
11
---dt=-.
0 t 4
Use this to show that
00
sin 2 t
1
11
-.,-dt = -.
o t- 2

Exercise 3.7.7. Use the fonnula sinx = 2sin ~cos~ repeatedly and show

sinx
-- = cos x- x
. cos - . cos - ...
x
X 2 4 8
This infinite product fonnula was discovered by Euler. Using the relation cos ~
J (I + cos x) /2 and choosing x = 11/2. obtain from this the formula

-=-·
11 2 2 2
This is called Viete 's product formula. Note that this formula allows us to obtain an ap-
proximate value of rr by repeatedly using four basic operations of arithmetic (addition,
multiplication, division, and square root extraction) all applied to a single number 2. Use
a pocket calculator to see what approximate value of 11 is obtained by taking the first ten
terms of this product.

3.8 The Gibbs phenomenon

We have seen that if f is piecewise C 1, then its Fourier series converges at each point
where f is continuous, and at a point of discontinuity the series converges to the average
value off at that point. However, in a neighbourhood of a discontinuity the convergence of
the series is not uniform and the partial sums of the series always overestimate the function
by about 18%. This is called the Gibbs phenomenon.
 ODDS AND ENDS 71

. Consider
. the function of the Example 3.3.5 and its odd extension having the Fourier ~

sme senes

-2 L<
T(
co I
-I )"+I_ sinmrx.
ll=l /l

For each N consider the partial sums SN(f; x) at the points x I - 1/ N and x
-1 + 1/ N. We have

SN (f: I - -NI) = -2L T(

N

ll=l
(-1)"+1
II
( nrr
sin n rr - - )
N

= 2
rr
t ~sin
11=1 11
nrr.
N

=2
rr
t
11=1
sin (nrr/ N) . ~-
nrr IN N

This last sum is a Riemann sum for the integral

rr -sindr t .
!o0 t

So we have, using Exercise 3.7.4,

lim SN ( f ; l - -I ) =.:.? !orr -dt>l.l1.

sin r
N-co N rr o t

By the same argument one can see that

lim SN(f;-1+_!_)<-1.17.
N~-oc N

The values of f at the points x = I- and x = -I+ are I and -1 respectively. However
the sums of the Fourier series o\'ershoot these limits.
If such calculations trouble you, you might find inspiration from the history of the dis-
covery of the Gibbs phenomenon.
You will have heard of A. Michelson, one of the greatest experimental physicists and
an extraordinary designer and builder of equipment to do the experiments. He is known for
his accurate measurement of the speed of light and for the Michelson-Morley experiments
to detect ether. He also designed and built a machine to calculate Fourier series. To test
his machine he fed into it the first 80 Fourier coefficients of the function f we have used
above. He was surprised that he did not get back the original function but the machine
seemed to add two little peaks near the discontinuities roughly as shown below.
72 FOURIER SERIES

The effect of increasing n was to move these peaks closer and closer to ±I but not to
diminish their size as Michelson expected. After verifying his observation by hand calcu-
lations he wrote a letter to the magazine Nawre in 1898 expressing his doubts that ..a real
discontinuity [in f] can replace a sum of continuous curves [S, (f: 0)]." Gibbs-another
great physicist who was one of the founders of modern thermodynamics-replied to the
letter and clarified the matter.

Exercise 3.8.1. You can clarify your understanding of uniform convergence by reconciling
the Gibbs phenomenon with Theorem 2.3.19 on convergence of Fourier series of piecewise
C 1 functions.

Exercise 3.8.2. We have demonstrated the existence of the Gibbs phenomenon for a par-
ticular function. Show that this must occur near the jumps of any piecewise C 1 function.

FIGURE7

3.9 Exercises
The common theme of the exercises in this section is posirh•ity. The numbers a,, b, and c,
in Exercises 3.9.1-3.9.4 are the Fourier coefficients associated with f as in (3.5) and (3.6).
(The index n runs over all nonnegative integers for a,, over all positive integers forb,, and
over all integers for c,.)

Exercise 3.9.1. Let f(r) > 0. Show that for alln, lc,l <co, Ia, I < a0 , and lb,l < a0 .

Exercise 3.9.2. Let f be an odd function such that f(t) :=: 0 for 0 < r < rr. Show that
lb,l < nb 1 for alln.

Exercise 3.9.3. Let f be a monotonically decreasing function on the interval (0, rr). Show
that b, > 0 for alln.
ODDS AND ENDS 73

Exercise 3.9.4. Let f be a convex function on (0, 2rr ). Show that a, :=: 0 for n = I. 2, ....
A doubly infinite sequence {c,} of complex numbers is said to be a positive definite
sequence if

(3.64)

for all positive integers N. and for all complex numbers .:o, .: 1, ••• , .:N _ 1• This is equiva-
lent to saying that for every N theN x N matrix

co C-(N-1)

Cl CO C-1

co (3.65)

CN-l CN-2 co
is positive (semi) definite. (This matrix has a special form; each of its diagonals is constant.
Such a matrix is called a Toeplitz matrix.) Clearly, if this matrix is positive definite, then

-
c_, = c,. co> 0, lc,l < co.

Exercise 3.9.5. Let f E C(T) and let {c,} be the Fourier coefficients of f.

(i) Iff (t) > 0 show that the sequence {c,} is positive definite.
(ii) Conversely. suppose {c,} is positive definite. Show that for all N

L Cr-si(r-s)t > 0.
O~r.s~N-l

This implies that the Fejer sums

aN(f; t) = L
N-l

11=-(N-J)
(
1-
I ') c,e i
11
N Ill
(3.66)

are nonnegative. Hence, by Fejer's Theorem (2.2.5), /(1) > 0.

This statement is a part of a general theorem called Herglot.: 's Theorem which says that
every positive definite sequence is the sequence of Fourier coefficients associated with a
positive measure 11; i.e.,
TT
c, =
i-TT
e -ill I dp,(t). (3.67)

(This can be proved along the lines of Exercise 3.9.5. Given c,,let /N(t) be the sequence
of functions defined by the expression on the right-hand side of (3.66). Then {fN} is a
family of nonnegative continuous functions with the property

-
1 Jj( /N(t) dt =co.
2rr -rr
 14 FOURIER SERIES

This is a kind of compactness condition from which we can derive (3.67) using standard
theorems of Functiona1 Analysis.)

Exercise 3.9.6. A linear operator A on the space C(T) is called positi\le if it maps non-
negative functions to nonnegative functions. (The elements of C(T) are complex-valued
continuous functions on T). Let go, g1, g2 be the three functions

go(t) = I, g 1(r) = sin t,

Koro"kin 's Theorem is the statement: If {A,} is a sequence of positive linear operators on
C(T) such that as n ~ oo,

A,gi ~ gi uniformly, for j = 0, 1, 2,

then

A,f ~ f uniformly for all f E C(T).

A proof of this theorem is sketched below. Fill in the details.

(i) It is enough to prove the assertion for all real-valued functions f

(ii) Fix s E T and let g5 (1) = sin2 [(t - s)/2]. Note that g5 (t) = go(t) - g, (s)gl (t) -
g2(s)g2(1).
(iii) Let M = sup Ifl. Let e > 0, and let D be such that 1/(1) - f(s)l < e whenever
It- sl <D. Note that g5 (t) > 4D 2 /rr 2 , whenever It- sl >D. Thus

Mrr-
.,
1/(1) - f(s)l < e + 2D2 gs(t).

(iv) Put K = Mrr 2 /2D 2 , a constant that depends on f and e. The last inequality can be
written as

an inequality between functions (in the pointwise sense). Since the operators A, are
linear and order preserving, this gives

I(A,f)(t)- f(s)(A,go)(t)l < e(A,go)(t) + K(A,g5 )(1) (3.68)

for all r.
(v) Choose t =sin (3.68) and let n ~ oo. Note (A,go)(s) ~ I, and by the observation
in (ii), (A,gs)(s) ~ 0. The convergence is uniform overs. So from (3.68) we see
that (A,f)(s) ~ f(s) unifonnly.

Exercise 3.9.7. Use Korovkin's Theorem to give another proof ofFejcr's Theorem. (Hint:
The Fcjcr sums A,f = a,(j) define positive operators on C(T).)
ODDS AND ENDS 75

Exercise 3.9.8. There is a Korovkin Theorem for the space C[a, bj where [a. b) is any in-
terval. The three functions go. g 1. g2 now are g,(r) = 1, g 1(/) = r. g2 (I) = r 2. Fonnulate
and prove the assertion in this case.

3.10 A historical digression

The series (3.39) is a famous one. The mathematician brothers Jakob and Johann Bernoulli
had discovered in 1689 a simple argument to show that the harmonic series L I1n is
divergent. This was contrary to the belief, held earlier. that a series L X 11 with positive
tern1s decreasing to zero converges. Among other series they tried to sum was the series
L II n'2 They could see that this series is convergent to a sum less than 2 but did not
succeed in finding the sum. This problem became famous as the ··Basel Problem.. after the
Swiss town Basel where the Bernoullis lived.
Solving the Basel problem was one of the early triumphs of L. Euler (1707-1783) who
went on to become one of the most prolific and versatile mathematicians of all time. Euler
was a student of Jakob and Johann. and a friend of Johann's sons Daniel and Nicolaus (all
mathematicians-there were more in the family). It is worthwhile to recall. very briefly.
some of the stages in Euler's attack on this problem as they are illustrative of a great
scientist's working... .
In 1731 Euler proved the fornmla

., ""- I
{(2) = (log2)- + 2"'"' --.,.
~ 211 n-
ll=l

using calculus and infinite series expansions of log( I - x). In this one can substitute
00 1
o = "'"'-.
Joa2 ~ n211
ll=l

The factors 211 occurring here make the series converge very rapidly. Using this he found
that

{(2) = 1.644934 ....

This number did not look familiar. More calculations of this kind followed. The proof of
(3.39), however, came through another route: Newton's theorem on roots of polynomials
generalized to ..infinite degree polynomials."
Let p(x) be a polynomial of degree n and suppose p(O) = I. Suppose we know the n
numbers a 1• a 2 , ••• , a 11 are roots of p. Then we can write

p(x) = (1 -_::__) (1- _::__) ···(1 -_::__).

a1 a2 a11
(3.69)

Euler considered the real function

• ") 5
smx x- x
f(x) = - = I - -3! +- · ·· . (3.70)
.r 5!
76 FOURIER SERIES

noticed f(O) = I, and f(x) = 0 precisely when x = mr, n E Z. In analogy with (3.69)
he factored (3.70) as

f (X) = [( 1- ~ ) (1 + ; ) ] [ ( 1 - ~: ) (I + ~: ) ] ···
=
( .,)( ")
1---
rr2
r-
1-~
r-
4rr-
...

He then multiplied out this infinite product to w1ite

I I I ) ., 4
f(x) = 1 - ( ?
rr-
+ -,
4rr-
+ -.,
9rr-
+ · · · x- + (· · ·)x + · · · . (3.71)

Now comparing coefficients of x 2 in (3.70) and (3.71) he saw that

I 1 I
- =?"'"" --:;.
3! rr- ~ n-

This is the remarkable fonnula (3.39). The answer confinned the numerical calculations
made earlier by Euler and others.
Several questions arise when one sees such a calculation. Can one do to an infinite
series whatever one does to a polynomial? One knows all the real roots off (x ), but are
there other roots? What is the meaning of an infinite product? The function g(.r) =ex f(x)
has the same roots as f(x). Can both be represented by the same fonnula?
Euler certainly was aware of all these problems and spent several years resolving them.
This led to a more rigorous proof of the product expansion (3.33). These days infinite
product expansions for entire functions (and meromorphic functions) are routinely used
in analysis. (See the Weierstrass Fact01isarion Theorem and the Mittag-Leffler Theorem in
L. V. Ahlfors, Complex Analysis.)
Euler found great delight in coaxing out of equation (3.71) more and more complicated
sums. The connection with Bernoulli numbers (introduced by Jakob Bernoulli in his fa-
mous work Ars Conjectandi) became apparent and he proved a general fonnula for {(2k)
in tenns of these numbers.
You may enjoy reading two books by W. Dunham. Chapters 8 and 9 of his book Journey
Tluvugh Genius are titled The Bemoullis and the Harmonic Series, and The Extraordinary
Sums of Leonhard Euler. The other book Euler: The Master of Us All is a short elementary
account of some of Euler's work. Chapter 3 of A. Weil. Number Theory: An Approach
Tluvugh His tot)' .from Hammurapi to Legendre is devoted to Euler. This book by a great
twentieth century mathematician is an excellent historical introduction to a substantial part
of mathematics.
The proof of (3.39) that we have given is one of several known now. Like Euler's proof
this uses a general method that leads to several other formulas. A more special proof given
by Euler is outlined in the following exercise:
ODDS AND ENDS 77
Exercise 3.10.1.
(i) Show that

1
-(sin- 1 x) 2 =
[,.t sin- 1 t
dt.
2 o J1 - t2

(ii) Use the integral

to expand sin- 1 x in a power series. Substitute this in the integral in (i) and integrate
term by term. Use the recurrence relation
1 11 2
t + + 1 [,' t
dt = - -
ll
11

dt
[ , --;::===::;;::
o J1 - t 2 n + 2 o J1 - t2

starting with

{ I --;::=t::::::;;: dt = I.
lo J1- r 2

(iii) The result of this calculation is a power series expansion for the function ~(sin -I x )2 .
Put x = 1 in this fonnula to obtain the sum (3.38).

Exercise 3.10.2. Show that

oo I co (n - I )2
L
11=1
-ll2 = 3 L -<-2,-l>-!-
11=1

[Hint: Use the power series for (sin- 1 x) 2 .]

One of the reasons for interest in the zeta function is the intimate connection it has with
prime numbers.

Exercise 3.1 0.3. For s > 1. let

col
{(s) = ~-.
~ ns
n=l

Show that

{(s) = n p
I
1 - p-S •

where, in the product on the right. p varies over all prime numbers. Among other things
this shows that there are infinitely many prime numbers. This fommla was discovered by
Euler.
78 FOURIER SERIES

Let us demonstrate an interesting application of it that links prime factors, probability,

and the number rr. A naive and intuitive idea of probability is adequate for our discussion.
A natural number picked ..at random" is as likely to be even as odd. We express this by
saying that the probability of a natural number being even (or odd) is 112. In the same way,
the probability of a natural number picked at random being a multiple of k is 1I k (as such
multiples occur at jumps of length k in the sequence of natural numbers).
Pick up a natural number n at random and factor it into primes. What is the probability
that no prime is repeated in this factoring? This happens if n is not a multiple of p 2 for
any prime p. As we have seen, the probability of this for any given prime p is 1 - 1I p 2 •
The probability of all these ..events" happening simultaneously is the product of these
individual probabilities, i.e., nr(l - 11 p 2). We have seen that this product is equal to

1 6
- - - -., .
{(2) rr-

Exercise 3.10.4. Think of a plausible argument that shows that the probability that two
natural numbers picked at random are coprime is 61rr 2 . Generalise this statement to k
numbers.

The fonnulas of Viete and Wallis (Exercises 3.7.7 and 3.5.3) were discovered in 1593
and 1655, respectively. Thus they preceded the work of Euler ( 1707-1783). Of course, the
arguments ofViete and Wallis were different from the ones given here. Since both fonnulas
give expressions for 2lrr. one may wonder whether there is a single expression that unites
them. There is one, and it was found quite recently. The curious reader should see the
paper T. Osler, The union of Vieta's and Wallis's product for rr, American Mathematical
Monthly. Volume 106. October, 1999, pp. 774--776.
A final tidbit. The symbol rr seems to have been first used by William Jones in 1706.
For thirty years it was not used again till Euler used it in a treatise in 1736. It won general
acceptance when Euler used it again in his famous book Introduction to the Analysis of the
Infinite published in 1748. There are a few books devoted exclusively to the number rr. A
recent one is J. Arndt and C. Haenel, rr Unleashed, Springer, 2001.
Convergence in L2 and L1

In Chapter 2 we saw that the Fourier series of a continuous function on T may not con-
verge at every point. But under weaker notions of convergence (like Cesaro convergence)
the series does converge. Another notion of convergence is that of convergence in the mean
square or L2 convergence. However, this notion is in some sense even more natural than
pointwise convergence for Fourier series. We will now study this and the related L 1 con-
vergence. We assume that the reader is familiar with basic properties of Hilbert space and
the spaces L 1 and L 2 • Some of them are quickly recalled below in the form of exercises.

4.1 L2 convergence of Fourier series

Let 1t be a Hilbert space with inner product (... ) and norm II II· Special examples of
interest to us are:

1. The space /2 of all sequences of complex numbers x = (x1. x2 •.. .) such that

The inner product in this space is defined as

00

(x. y) = L X11Y11
11=1

and the associated norm is

co ) 1/2
llxll2 = (
L lx11l
11=1
2

79
80 FOURIER SERIES

We need also the space of doubly infinite sequences {x, l~-oo satisfying
c-:
L
11=-X
lx,l2 < oo.

We denote this space by /2(2). It is easy to see that /2 and /2(2) are isomorphic Hilbert
spaces.
2. For any bounded interval I on the real line, the space L2(1) consists of all Lebesgue
2
J
measurable functions f on I such that 1 1!1 < oo. The inner product on this space is
defined as

(f. g)=- If -
Ill I
f(x)g(x)d x,

where Ill denotes the length of /. The associated norm is

1 { 2 ) 1/2
II !112 = ( 1/1 lr lf(x)l dx

(The factor 1/111 is inserted to make some calculations simpler.)

3. The space L2(IR) consists of all Lebesgue measurable functions on the real line IR such
that fi?.I/1 2 < oo. Here the inner product and the norm are given as

(f. g) = { f(x)g(x) dx.

liP.
1/2
2
11!112 = (£.1f(x)l dx)

Of course. when we talk of measurable functions. we identify functions that are equal
almost everywhere.
The space L2[-rr. rr] can be identified with L2(T).
We are interested also in the space L 1(/) which consists of all integrable functions
on /,i.e.,

L1(l) = {t: f lf(x)ldx < oo}

When I is a bounded interval. we define. for f E L 1( /).
1
11!111=- {lf<x)ld x.
Ill l1
For f in the space L 1(!R) we define

11!111 = ~ lf<x>l dx.

The spaces L 1(/) and L 1(!R} arc Banach spaces with this nonn. Neither of them is
a Hilbert space. (You can prove this by showing that the nonn does not satisfy the
"parallelogram law" which a Hilbert space nonn must satisfy).
CONVERGENCE IN L2 AND L1 81

Exercise 4.1.1. Show that the nonn and the inner product in any Hilbert space satisfy the
Cauchy-Schwar~ inequality:

l(x. y)l ~ llxll IIYII·

Exercise 4.1.2. Use this inequality to show that for every bounded interval/,

Show that this inclusion is proper.

Exercise 4.1.3. Show that neither of L2(IR) and L 1(IR) is contained in the other.

Exercise 4.1..1. Show that continuous functions are a dense subset of L2 and of L l· Func-
tions of class C 00 also are dense in L2 and in L l·
Recall that a sequence {e,} in a Hilbert space 1-l is called orthogonal if (e,, e111 ) = 0 for
n f= m, and orthonormal if in addition lie, II= I. Given any vector x in 1-llet (x, e,) =a,.
The numbers a, are called the Fourier coefficients of x with respect to the orthonormal
system {e, }.

Exercise -1.1.5. Show that in the Hilbert space L2(T) the sequence e,, defined as

e,(x) = ei"x. n E Z.

is an orthonormal sequence. Note that here the Fourier coefficients are

(f, e,) = -
1 iTT ..
{(x)e-"1.1; dx = f(n).
A

2rr -:r

So the general results on orthononnal sets in Hilbert spaces are applicable here.

Exercise -1.1.6. (Bessel's inequality) Prove that if {e,} is any orthononnal system in a
Hilbert space 1-l and a, the Fourier coefficients of a vector x with respect to this system,
then

Therefore, the sequence {a,} belongs to the space l2. For f E L2(T) this gives

L lf<n>l
II
2
< 11!11 2•

Exercise 4.1.7. Let a, be the Fourier coefficients of a vector x with respect to an orthonor-
mal system {e,}. Let b, be any other sequence of numbers. Show that

N N
x- La,e, < x- Lb,e,
11=1 ll=l
82 FOURIER SERIES

for all integers N. Further. this inequality is an equality if and only if a, = b, for all
n = I. 2..... N. For the space L2( T) this means that among all trigonometric polynomials
of degree N. the partial sum SN(f) is the closest approximation to a given function fin
the norm of L2(T).
An orthonom1al system {e,} in a Hilbert space 11 is called complete if finite linear
combinations of {e,} are dense in 11, i.e., given x E 11 and e > 0 there exists an integer N
and numbers ai ... ,a;v such that
N
.l • - ~a
" " ' I e 11
11 <e .
ll=l

Note that, by Exercise 4.1.7, the best choice for such numbers. if they exist. is the Fourier
coefficients a,. A complete orthononnal system is called a basis for 11. The space 11 is
separable if and only if it has a countable basis.

Exercise -1.1.8. (Plmzclzerel's Theorem) Show that an orthonormal system {e,} is a basis
in 11 if and only if the Fourier coefficients a, of every vector x with respect to e, satisfy
the equality
""' Ia, I-=
~ ., llxll-..,
II

i.e., Bessel's inequality becomes an equality.

Theorem 4.1.9. The system e,(x) = ei"x is a basis for L 2 (T).

Proof Let f E L2(T). By Exercise ..J.l.4, for each e > 0. there exists g E C 1(T) such
that ll.f- glb < ~·By Dirichlet's Theorem there exists anN such that
e
sup lg(x)- SN(g; x>l < -.
')

Denote SN(g; x) by SN. Then

Hence

But SN is a linear combination of the functions e,.

•
Exercise .J.l.JO. Show that for f E L 2 (T),
N
lim
N-..x
f- L /(n)e,
11=-N
-0
- .
2
:"v

11!11~ = L IJ\n>l
11=-00
2
CONVERGENCE IN L2 AND L1 83

Thus the Fourier series L j (n )einx converges to f in the sense of convergence with respect
to the norm of L2(T). In particular. this is true for all continuous functions on T. This
explains our remark at the beginning of the chapter.

Exercise 4.1.11. Let (a11 } be any sequence in l2. Let 1-l be a Hilbert space with an or-
thonormal basis {en}. Show that there exists a vector x in 1-l whose Fourier coefficients
with respect to e" are a". Hint: Show that the series L a11 e11 converges in 11.
In particular. given any sequence {a 11 }~_ 00 in l2(Z) we can find a function f in L2(T)
such that

j(n) =a 11 •

Exercise 4.1.12. (Riesz-Fischer Theorem) Show that the spaces L2(T) and l2 are iso-
morphic: the map that sends an element f of L2(T) to the sequence consisting of its
Fourier coefficients is an isomorphism. (This is a very important fact useful in several con-
texts. For example. in Quantum Mechanics the equivalence between the ..Wave Mechan-
ics" and the .. Matrix Mechanics" approaches is based on this fact. See the classic book.
J. von Neumann. fv/athematical Foundations of Quantum Mechanics.)

Exercise 4.1.13. (Parseval's Relations) Show that if f. g E L2(T). then

-
I ITT f(xfg(x) dx = L
(X) A -

f(n)g(n).
2rr -rr n=-oo

Prove a general version of this for any Hilbert space.

Exercise 4.1.14. Use the Weierstrass approximation theorem (Exercise 3.1.4) and Exer-
cise 4.1.4 to show that the polynomial functions are dense in the Hilbert space L2[ -1. I].
Note that a polynomial is a finite linear combination of the functions x". n > 0. Obtain
an orthonormal basis of L2 f-I. I] by applying the familiar Gram-Schmidt process to the
functions x". The functions you get as a result are. up to constant multiples.

Po(x) = 1.
P,(x) = x.
3 ., 1
P2(x) = 2x-- 2'
5 3 3
PJ(X) = ?X - ?X.
- -
I d" ., II
P11 (X) = - - ( . C -1) .
2"n! dx"
These are called the Legendre Polynomials. Every function in L2[ -1. I] can be expanded
with respect to the basis formed by these functions. This is called the Fourier-Legendre
84 FOURIER SERIES

series. (There are several other orthonom1al systems for the space L2 that are useful in the
study of differential equations of physics).

Exercise 4.1.15. Let {e,} be an orthononnal system in a Hilbert space 11. Show that the
following conditions are equivalent:
(i) {e,} is a basis.
(ii) if x E 11 is such that (x. e,) = 0 for alln. then x = 0.
(iii) llx f = 2::,1 (x. e,) 12 for all x E 11.

These results are very useful in different contexts. Here are some examples.

Exercise4.1.16. Let f E C 1(T). Recall that then inj(n) = ['(n). (See Exercise 2.3.11).
Use tltis and the Cauchy-Schwarz and the Bessel inequalities to show that. for 0 < N <
N' < oo,

ISN(f; 0)- SN'(f; 0)1 < L li<n>l

l11l>N
1
= L -
llll>N In I
1]\n)l

c I
< Nl/2 II! 112-
This shows that SN(f; 0) converges to f(O) uniformly at the rate I/ N 112 as N-..:;. oo.
By the same argument. show that iff E Ck(T). then SN(f; 0) converges to {(0)
unifmmly at the rate I 1Nk-l/ 2.

Exercise 4.1.17. In Exercise 2.3.12 we saw that iff E Ck(T). then j(n) = 0(1/nk).
Show that iff E L 1(T) and j(n) = 0(1/nk) for some integer k > 2. then fisk - 2
times continuously differentiable. Show that no more can be concluded.
Hint: L j(n)inei"x converges uniformly.

Exercise 4.1.18. A sequence {x, l~-oo is called rapidly decreasing if lnlk x, converges
to zero as In I -..:;. oo. for all k > 0. Construct an example of such a sequence. Show that
f E C 00 (T) if and only if the sequence j(n) is rapidly decreasing.
Exercise 4.1.19. Let f have the Fourier series
00
ao
2 "'""' cos nO+ b,
+~(a, sin nO).
II= I

(See (3.5) in Chapter 3). Show that

1T ., co
I
:rr
1
_ lfl =
1T
2 Qi}
2 + L<a~ +b~).
ll=l
., .,

Derive a corresponding result when f is a function with an arbitrary period.

CONVERGENCE IN L2 AND L1 85

Exercise 4.1.20. Use this version of Plancherel's Theorem to prove

Hint: Consider the Fourier series of the function in Example 3.3.5.

Exercise 4.1.21. Derive more fonnulas of this kind by applying Plancherel's Theorem to
functions studied in Section 3.3.

Exercise 4.1.22. Let f be a C 1 function on [0. rr] with /(0) = {(rr) = 0. Show that

By a change of variables. show that iff is a C 1 function on any interval [a. b] with f(a) =
{(b) = 0. then

b a)21b
la
1/1 2 <
(b
-
11
2
a
If'(!

This is called lVirtinger's inequality.

Show that these inequalities are sharp. Hint: Let f(x) =sin x.

4.2 Fourier coefficients of L1 functions

The Fourier coefficients

f(n) = -1 ~TT j(t)e-1111

•
dt
2rr -rr

are defined whenever the integral on the right-hand side makes sense. This is so whenever
f E L 1(T). The spaces C(T) and L2(T) we have considered earlier are subs paces of
L 1(T). For L2(T) we have obtained a complete characterization of the Fourier coefficients:
we know that f E L2(T) iff the sequence j(n) is in /2. and further the map taking f to the
sequence {j(n)} is an isomorphism of Hilbert spaces. For Lj(T) the situation is far less
satisfactory. as we will see in this section. One of the more attractive features of L 1(T)
is that it is a "convolution algebra" and there is a very neat theorem relating the Fourier
coefficients of f * g with those of f and g. Many of our results for L 1 can be deduced
from the ones we have already derived for continuous functions.
For f E C(T) define

11/lloo =sup 1/(t)l.

tET

Then note

II/III =
1 17( lf(t>l dt < 11/lloo·
-rr (4.1)
211
86 FOURIER SERIES

So if j;, ~ fin the space C(T). i.e.. if the sequence j;, of continuous functions converges
uniformly to f. then .f, ~ fin L 1(T) also. i.e.• IIJ,, - !II 1 ~ 0.
For}: g E L 1(T) define. as before.

(f * g)(t) = frr f(t - x)g(x) dx.

-TT

A routine argument using Fubini's Theorem shows that f *g is well defined for almost all
t. and is in L 1(T). It is perhaps worthwhile to go through this argument in detail. though it
is somewhat dull.
Let I. J be any two (finite or infinite) intervals and let I x ./ be their product. The
theorems ofFubini and Tonnelli relate the double integral of a measurable function f(x. y)
with respect to the product measure on I x J to iterated one-variable integrals.

Fubini's Theorem. Let f E L1 (I x J), i.e., let

{{ lf(x • .v>l dx dy < oo.

JlrxJ
Then for almost all x E I.

i lf(x. y)j dy < oo.

i.e., the jimction J:r:(Y) = f(x. y) is an integrable function of y for almost all x. Furthe1;
the jimction

i f(x. y)dy

is an integrable fimction of x, and

ffxl f(x, y)dxdy = f [i l

f(x. y)dy dx.

A corresponding statement with the roles ofx andy interchanged is true.

Tonelli's Theorem. Let f be a IWilllegath•e measurable function on I x J. If one ~f the

integrals

f [i f(x,y)dy]dx,

i [! f(x. )') dx] dy

is finite, then so is the otlte1; and f E L 1(I x J).

CONVERGENCE IN L2 AND L1 87

Hence (by Fubini's Tlteorem)

flxJ f(.r,y)d.rdy = 1[.l f(x.y)dy]d.r

= .£[1t<x.y)dx]dy.
(Fubini's Theorem says that if f(x, y) is an integrable function on I x J. then its
integral can be evaluated in either of the three ways-as an integral with respect to the area
measure or as an iterated integral in two different ways. Tonelli's Theorem says that iff
is nonnegative. then even the hypothesis of integrability is unnecessary. The three integrals
are either all finite or all infinite.)
Let us return to convolutions now.

Theorem 4.2.1. Let f, g, E L l(T). Then the integral defining <f * g)(t) exists absollltely
for almost all t, is in L 1(T), and
(4.2)

Proof Since f is a periodic function,

~~ lf(t -x)jdl = ~~ lf(tlldl = 2rrllfll1

for all x E T. Hence

1"
-rr lf(t -x)g(.t)ldt = lg(x)l 1"
-rr lf(t -x)ldt = lg(x)l2rrllflll·
and therefore,

I:[!~ lf(l- x)g(x)j d1 Jdx = 4rr 2 11flh 118111·

So, by Tonelli's theorem f<t - x)g(x) is an integrable function on the rectangle T x T.

r:
By Fubini's theorem

(f * g)(l) = f(t - .t)g(x) d.r

is well defined for almost all t. We could put an arbitrary value 0 for f *g at the exceptional
set where it may not be defined. Now note that

2rr II f * g IIi = r: I(f * g)( I) Id I

= ~~ ~~ [(1 - x)g(x) dx d1 < ~~ ~~ lf(l- .r)g(.r)l d.t d1

=I: lg(x)j [!~ 1/(1- .r)j d1] dx
2
= 4rr 11flllllglll •
88 FOURIER SERIES

An immediate application of this is:

Theorem 4.2.2. Let f E L 1(T) and let a, (f) be the Cesaro sums of the Fourier series
of f. Then

lim II!- a,(f>lll = 0,

,_00
i.e., the Fourier series off is Cesaro swmnab/e to f in L 1 norm.

Proof Continuous functions are dense in L 1(T). So given £ > 0 there exists a continuous
function g such that II f - g II 1 < e. If F, denotes the Fejcr kernel, defined by (2.8), then
by Theorem 2.2.5 and the inequality (4.1) there exists N such that II g - g * F,, II 1 < £ for
alln > N. Hence, we have,

II f - f * F, II I < II f + II g - g * F,, II I + II g * F,
- g II I - f * F,, II I
< 2£ + II (g - !> * F,, II I .

But. by Theorem 4.2.1,

So

II f - f * F, II 1 < 2( I + rr )e for alln > N. •

Corollary 4.2.3. (The Riemann-Lebesgue Lemma) Iff E L 1(T), then
A

lim f(n) = 0.
lnl-oo

Proof Use Theorem 4.2.2 and the idea of the proof of Theorem 2.3.1.
•
Let co denote the space of sequences converging to 0. The Riemann-Lebesgue Lemma
says that the map f ~ {j (n)} is a map from L 1(T) into co. From results on continuous
functions obtained earlier, we know that this map is one-to-one. This map has a particularly
pleasing behaviour towards convolutions:

Theorem 4.2.4. Let f and g be in L 1(T). Then

- A

(f*g)(n) =2rrf(n)g(n) fora/ln.

 CONVERGENCE IN L2 AND L1 89

Pmof Once again. using Fubini 's Theorem we can write

(f- * g)(n) = - 1
2:rr
I (f * g)(t)e- .
1111
dt

= 2~ 11 rcr- dt d.t
.t)g(.t)e_;.,

= 2~ I I f(t- .r)e_;,,, __,,g(x)e-;"·' dt dx

1
= 2:rr I · I
f (t)e _,, d t g (x )e ·
_,,x d x
= 2:rr j(n)g(n).
(All integrals above are over the interval [-:rr, :rr ].)
•
Remark. Some books define f * g with a different normalisation as
<f * g)(t) =? 1 iTT f(t- x)g(x) d.Y.
-:rr -rr

With this definition you will get

(f
-

* g)(n) = f(n)g(n)A

and

A vector space X is called an algebra if its elements (x. y, :. etc.) can be multiplied
and the product x • y obeys the relations

x • (y • :) = (x • )') • ~.

(ax) • }' = a(x • y) for all scalars a.

Suppose X is a Banach space with nonn 11.11. If a product on X satisfies the inequality

II x • y II < II x II II y II for all x. )',

we say X is a Banach algebra.

The space C(T) (with f + g and f g defined as usual) is a Banach algebra with the
supremum nmm. The space M (n) of n x n matrices is a Banach algebra with the nonn
of A defined as HAll = sup{IIAull : u E C". Hull = 1}. The multiplication in C(T) is
commutative, that in M (n) is not. Our discussion shows that L 1(T) is a Banach algebra
with multiplication of f and g defined by convolution f * g (defined in such a way that
II f * g II I < II f II "' g II I ) .
90 FOURIER SERIES

Exercise 4.2.5.
(i) Show that there does not exist any function f in L 1(T) such that f * g = g for all
g E Lt (T).
(ii) Show that there does not exist any continuous function f E C(T) such that f *g = g
for all g E C(T).
(iii) Let f E L 1(T) and g E Ck (T). Show that f * g is in Ck (T), and its kth derivative is
(f * g)<k) = f * g<k>.

Thus neither the space of continuous functions, nor the space of integrable functions has
an identity with respect to convolution. (The constant function I is an identity for C(T)
with the usual multiplication, the identity matrix is an identity for M (n).)
We have seen that iff E L2(T). then j(n) not only goes to zero as In I ~ oo, but is
square summable. However, for f E L 1(T) nothing more than j(n) ~ 0 can be said in
general. This is proved in Theorem 4.2.1 0 below.

Exercise 4.2.6. Let f be absolutely continuous and let f' E L2(T). Show that

[Hint: Use the Cauchy-Schwarz Inequality.]

Note that in particular this implies that j(n) is in /1 (Z).
Let {a,}~ 0 be a sequence of real numbers. Let D.a, =a,+ I -a,. D.2a, = D.(D.a,) =
a11 +2 - 2a11 +I +a,. Note that {a,} is monotonically increasing if and only if D.a, > 0 for
alln.

Deji11ition 4.2. 7. The sequence {a,} is said to be con\'ex if D. 2a, > 0 for alln.

Proposition 4.2.8. Suppose the sequence {a, }~ 0 is conl'ex and bounded. Then
(i) {a,} is monotonically decreasing and convergent,
(ii) limn D.a, = 0,
(iii) L~o(n + l)D. 2a, = ao -lima,.

Proof If D.a, > 0 for any n, then by the given convexity, D.a, > D.a, > 0 for allm > n.
But then {a,} cannot be bounded. So D.a, < 0 for alln. This proves (i). Note that
II

L D.ak =a,+ I - ao.

k=O

Thus the series L( -D.ak) is convergent and its terms are positive and decreasing. This
proves (ii). We have
N
L<n + l)D. 2a, = ao- aN+ I+ (N + l)D.aN+l·
11=0

This proves (iii).

•
 CONVERGENCE IN L2 AND L1 91

Theorem 4.2.9. Let {a, l~-oo be a sequence such that

(i) a, > Ofor al/n,
(ii) a, =a_,,
(iii) a, ~ 0 as n ____:;. oo,
(iv) {a, }~ 0 is com'e.Y.

Then there exists a nonnegative fimction f E L 1( T) such that a, = f (n) for a/ln.
Proof Let F, be the Fejer kernel and
00

[(t) = 2rr L<n + l)~ a,F,,+ 1 (t).

2

11=0

Since IIF,,II = l/2rr for alln, this series converges in L 1. So f E Lj(T). Note that
f (t) > 0 since all the terms involved are positive. For each integer k. we have
00

i<k> = 2rr L<n + l)~ a,F,+ 1 <k).

2

11=0

Since

F,+l(t) = -
I II

2rr j=-11
.
L (
I---
n+I
Ul ) e'l',
..

we have

~1-lkl/(n+l) forlkl<n.
2rr F,,+!(k) =
A

0 for lkl > n.

Hence

f(k) = ~ ~ lkl-) = alkl·

A
~ (n
"=lkl
+ l)~-a, (
I--
n+I
•
Exercise 4.2.10. There exists a sequence of nonnegative real numbers that is convex and
converges to zero more slowly than any given sequence; i.e .. if {b,} is any sequence con-
verging to zero. then there exists a convex sequence {a,} such that lh,l < a, and a, ____:;. 0.
[Hint: Think of functions instead of sequences.]

We have shown above that a sequence of nonnegative real numbers can be the sequence
of coefficients of a Fourier cosine series and converge to zero at any rate. The situation is
different for a sine series:

Theorem 4.2.11. Let f E L !( T) and let

(i) /(n) > O.for al/n > 0,

(ii) j(n) =- j(-n)foral/n.
92 FOURIER SERIES

Then

'~-
"""1 f(n) < oo.
A

wf.Oil

Proof Let

F(x) = r: f(t)dt, -T! <X< T!.

Then F is absolutely continuous and F' = f, a.e. Hence

A 1 A

F(n) = -. f(n) for alln =ft 0.

Ill

Now apply Theorem 2.2.5 to the function Fat the point 0 = 0 to get

F(O) = F(O)
A

+ lim2~
~ (

1-
n ) j (n)
-.-.
N-co ll=l N + 1 Ill

So
N A

lim '""" ( 1 -
~ n ) -f(n) = -i ( F(O) - F(O) . A )

N-oo ll=l N +1 n 2

Thus the series L j(n)/n is Cesaro summable; and since its terms are nonnegative it must
be convergent. •

Corollary 4.2.12. If a, are nonnegati,•e rea/numbers and if

00
'"'a,
~- = 00,
11=1
n

then the series L:~ 1 a, sin nt cannot be the Fourier series of anyfimction in L 1(T).

Exercise 4.2.13. Show that the series

00
L cosnt
11=-
., log n

is the Fourier series of some function in L 1(T), but the series

00 •
Lsmnt
11=-
.., logn

cannot be the Fourier series of any function in L 1(T).

So, what can be said about the class A of all sequences which are Fourier coefficients
of functions in Lt (T)? We have shown the following:
CONVERGENCE IN L2 AND L1 93

(i) A C co, the space of doubly infinite sequences going to zero at ±oo,
(ii) A# co,
(iii) A is an algebra. i.e., it is closed under addition and (pointwise) multiplication because
of the relations:

(f
-

+ g)(n) = A

/(n) + g(n).
<f--;g)(n) = 2rr j(n)g(n).

However, a complete description of the class A is not known. (In fact, it is believed that
there is no good description of A).
Some Applications

One test of the depth of a mathematical theory is the variety of its applications. We began
this study with a problem of heat conduction. The solution of this problem by Fourier
series gives rise to an elegant mathematical theory with several applications in diverse
areas. We have already seen some of them. We have proved the Weierstrass approximation
theorem in Chapters 1 and 3. This is a fundamental theorem useful in numerical analysis,
in approximation theory and in other branches of analysis. In Chapter 3 and 4 the Dirichlet
and Plancherel theorems were used to obtain the sums of some series and the values of
some integrals. Some more applications of Fourier series are described below.

5.1 An ergodic theorem and number theory

The ergodic principle in statistical mechanics states the following: the time average of
a mechanical qualltity should be the same as its phase average. Rather than explain the
general principle we will find it instructive to see how it operates in a special situation. The
model we describe is due to H. Weyl.
Consider the circle T =[ -rr. rr ). Let cp E ( -rr. rr) and define a map Rrp : T ~ T by

Rrp(O) = e+cpmod2rr. (5.1)

If T is thought of as the unit circle in the plane. then Rrp is a rotation by angle cp. We think
ofT as the phase space and the action of Rrp as dynamics on this space. If e = eo is a given
point on T. then its trajectol)' under this dynamics is the succession of points

eo= e.
e1= Rrpe = e + cp mod 2rr.
e2 = R~e = e + 2cp mod2rr,

95
96 FOURIER SERIES

Any continuous function f on T is thought of as a mechanical quantity. The time Q\'erage

off is defined as

l N-1
lim -
N-co N
L
k=O
f(Ok), (5.2)

and its phase average is defined as

- 1 ITT f(O)dO. (5.3)

2rr -TT

One can think of Oo moving to 0 1, fh, ... at successive time intervals. The sum in (5.2)
then gives the average value of f on the trajectory of f from time 0 to N. The integral
(5.3), on the other hand, gives the average value off over the phase space T.
An example of the ergodic principle is:

Theorem 5.1.1. (Weyl) If cp is an irrational multiple of2rr. then the quantities (5.2) and
(5.3) are equal.

Proof Given a continuous function f write WN (f) for the difference between (5.2) and
(5.3):

1 N-1 1 ITT
WN(f) = -
N k=O
L
f(Ok)--
2rr -TT
f(O)dO.

We want to show WN(f) ~ 0 as N ~ oo. The idea of the proof is to do this first for the
functions e, (0) = e;,o. then for trigonometric polynomials and then by Fejer's Theorem
for any continuous function f.
When n = 0, e, (0) = 1; and for f - 1 both (5.2) and (5.3) are equal to 1. So, in this
case WN = 0 for all N.
When n f:. 0 we can write

'w N (eII )I
l
= -
N
N-1
"'e''"(O+krp) -
~
k=O
_1_ I
2rr -TT
TT
ei"O dO

ei"O 1 _ ei11Ncp

N 1 - ei"cp
2 1
<-
- N 1 - eiucp

Note that ei"cp f:. 1 since cp is an inational multiple of 2rr. As N ~ oo, the right-hand side
of the inequality above goes to 0.
SOME APPLICATIONS 97

Now note that both (5.2) and (5.3) are linear functions of f. So

WN(p) ~ 0 as N ~ oo,

whenever pis an exponential polynomial.

This linearity also shows that if .f and g are continuous functions on T such that

sup lf(O)- g(O)I < e,

then

IWN(f)- WN(g)l < 2e, for all N.

Now use Fejer's Theorem to complete the proof. •

Exercise 5.1.2. Show that if cp is a rational multiple of 2rr, then (5.2) and (5.3) are not
always equal.

A remarkable consequence of Theorem 5.1.1 is Weyl's Equidistribution Theorem in

x
number theory. Let denote the fractional part of a real number x, i.e., 0 < < 1 and x-x x
is an integer. Weyl 's equidistribution theorem says that if x is irrational, then for large N
the fractional parts of x, 2x, ... , Nx are uniformly distributed over (0, 1). A precise state-
ment is:

Theorem 5.1.3. (Weyl) If x is an irrationalnumbe1; then for eve I}' subinterval [a, b] of
(0, 1),

lim _!_ card{k: 1 < k < N, (kx)"' E [a, b]} = b- a. (5.4)

N-co N

Here card A denotes the cardinality of a set A, and (kx)"' is the fractional part of the
number kx.

Proof By a simple change of variables the statement (5.4) is reduced to an equivalent

statement: if x E T and is an irrational multiple of 2rr, then for every subinterval [a, b] of
(-rr,rr),

1 b-a
lim - card{k: 1 < k < N, (kx)"' E [a, b]} = -,-. (5.5)
N-co N ... JT

Of course, here kx is again thought of as an element of T for all k.

We will prove (5.5) by applying Theorem 5.1.1. Let X[a,bJ denote the characteristic
function of [a, b ], i.e.,

1 if t E [a, b],
X[a.bJ(f) = { 0 otherwise.
98 FOURIER SERIES

This is a discontinuous function. Given e > 0 we can choose two functions f+ and!-
which are continuous and approximate X[a.bJ from inside and outside to within e, i.e ..

0 < f_(t) < X[a.bJ(t) < f+(t) < 1 for all t (5.6)

r:
and

(b- a)- e < J_(t) dt < !~ f+(l) dt < (b- a)+ e. (5.7)

Note that (5.6) implies

N N
L f- (<kxf"} < card{k: 1 < k < N, (kx) .... E [a, b]} < L f+ ((kx('). (5.8)
k=l k=l

By Theorem 5.1.1, there exists No such that for all N > No,

(5.9)

Now use (5.7), (5.8) and (5.9) to obtain (5.5). •

Notice that we have interpreted the process of picking up the fractional parts of
x, ?_x, 3x . ... as a ··d ynam1ca
. I process."

Exercise 5.1..1. Show that the statement of Theorem 5.1.3 is not true when x is a rational
number.

5.2 The isoperimetric problem

Among all simple closed plane curves with a given perimeter, which one encloses the
maximum area? <Here a ..simple'· curve means a curve that does not intersect itself.) This
is called the isoperimetric problem. The Greek founders of geometry proved that a circle of
perimeter L always encloses a larger area than any polygon with the same perimeter. This
result is attributed to Zenodorus (famous for "Zeno's paradox"). In 1841 Steiner proved that
among all simple closed plane curves with perimeter L. the circle encloses the maximum
area. A very elegant proof of this fact using Fourier series was given by Hurwitz. We will
give this proof here under the additional assumption that all curves that we consider arc
(piecewise) C 1•
Let C be any curve in the plane parametrised over f -:rr. rr ], i.e .. C is the range of a
continuous map (x{t). y(t)) from [-rr, rr] to IH!2. You can think of Cas the trajectory of a
particle moving in the plane from time -:rr to rr. We assume that the curve is

(i) closed, i.e .. (x( -rr), y( -rr)) = (x(rr). y(rr)).

(ii) simple, i.e., if -rr < s < t < rr then (x(s), y(s)) ;j= (x(t), y(t)),
(iii) smooth, i.e., x(t) and \'(t) both are C 1 functions,
SOME APPLICATIONS 99

(iv) of length L, i.e.,

It is convenient to choose the parametrisation in such a way that

L2
(x' (t)) 2 + (y' (t)) 2 = -.,. (5.10)
4rr~

In the picture of the moving particle this means that the particle moves with unifonn speed
while tracing the curve C. This involves no loss of generality in our problem.
If A is the area enclosed by such a curve, then

A=~~ x(t)y'(t)dt. (5.11)

Integrate (5.1 0) over [ -rr, rr] and use Plancherel's Theorem (Exercise 4.1.1 0). This gives

L2
-., = -I iTT [(x'(t)) 2 + (y'(t)) 2] dt
4rr- 2rr -rr

00

= L (lx'(n)l
11=-00
2
+ ly'(n)l 2 )
00

= L n 2 (1x(n)l 2 + ly(n)l )
2 (5.12)
11=-00

by Exercise 2.3.11.
Now note that, since x(t) and y(t) both are real functions, .t(n) and )'(ll) are complex
conjugates of x( -n) and)'( -n) respectively. So (5.12) can be written as
00
2
L = 8rr 2
L n (1x(n)l
2 2
+ ly(n)l 2 ). (5.13)
11=1

Similarly, using Parseval's relations (Exercise 4.1.13) we get from (5.11)

00

A = 2rr L
11=-00
.i(n)()"(n))

00

= -2rri L
11=-00
nx<n>.Y<n>
00
= 4rr L n Im(x(n))'(n)). (5.14)
II= I

Here Im denotes the imaginary part and the bar denotes complex conjugation.
From (5.13) and (5.14) we get
00 --
L 2 - 4rr A = 8rr 2 L{n 2 (1x(n)l 2 + l.v(n)l 2) - 2n Im.i(n)y(n)}. (5.15)
II= I
100 FOURIER SERIES

Now if .t(n) =a, + ifJ, and y(n) = y, + iD, are the respective real-imaginary decompo-
sitions. then the nth tenn of (5.15) can be written as

+ {3,2 + y,2 + D,)-

n 2 (a,2 2 _n(fJ,y, -a, D)
,., ,

= (n 2 - 1)(a,2 + {3,)2 + (a, + nD,) 2 + (/3, - n y, )2 ,

and this is nonnegative for each n. Hence from (5.15) we get

')
L- > 4rrA

and equality holds here if and only if

a, = {3, = y, = D = 0
11 for n > 2,

Hence, we have L2 = 4rr A if and only if

x(t) = x(O) + 2(at cost - fJI sin t),

y(t) = .VCO) + 2(/31 cost +at sin t),

or equivalently, if and only if

(x(t) -.t(0)) 2 + (y(t)- y(0)) 2 = 4(af + /3f) for all t.

This means that the curve C is a circle.

We have proved that among all smooth simple closed curves in the plane with a given
perimeter, the circle encloses the maximum area. With a little work the condition of
smoothness can be dropped. You can try proving this by assuming the following: if C
is a rectifiable simple closed curve in the plane with perimeter L and enclosed area A,
then there exists a sequence of smooth simple closed curves C, such that C, converges
uniformly to C; further if L, and A, are, respectively, the perimeter of C, and the area
enclosed by C,. then L, ~ L and A, ~ A.
The isoperimetric problem is believed to be the first extremal problem discussed in the
scientific literature. It is also known as Dido's Problem. The epic poem Aeneid by the
poet Vergil of ancient Rome tells a legend: Fleeing from persecution by her brothel; the
Phoenician princess Dido set off westward along the Mediterranean shore in search of a
ha\'en. A certain spot on the coast of what is now the bay of Ttmis caught her fan c)~ Dido
negotiated the sale of/and with the localleade1; Yarb. She asked for ''el)' little-as much as
could be "encircled with a hull's hide.'' Dido managed to persuade Yarb. and a deal was
struck. Dido then cut a hull's hide into narrow strips. tied them togethe1; and enclosed a
large tract of/and. On this land she built a fortress and, near it. the city ofCartlwge. There
she was fated to experience unrequited love and a martyr's death.
This quote is from Stories about Maxima and Minima by V.M. Tikhomirov. We recom-
mend this book, and Chapter VII of the classic What Is Mathematics? by R. Courant and
H. Robbins, where several extremal problems arising in mathematics, physics, astronomy,
and other areas are discussed.
SOME APPLICATIONS 101

5.3 The vibrating string

The solution by d' Alembert of the problem of the vibrating string was the beginning of the
development of Fourier series. The connection is explained in this section.
Just as we derived the Laplace equation for steady state heat conduction from a simple
law-the Newton law of cooling-we can derive the equation governing the vibrations of
a string using another simple law of Newton-the Second Law of Motion.
Imagine a thin. long, stretched elastic string. If the length is much greater than the
thickness we may think of it as a one-dimensional string stretched between points 0 and L
on the x-axis. If it is plucked in the y direction and released, it will start vibrating. We
assume that the tension of the string is high and the vibrations are small, all motion takes
place in the x-y plane and all points of the string move perpendicular to the x-axis. These
assumptions, no doubt, are simplistic but they describe the behaviour of a typical system-
such as a musical instrument-fairly accurately.
Let u(x, t) describe the profile of the string, i.e., u(x, t) is the displacement in the
y-direction at time t of a point which was originally at point x in the equilibrium posi-
tion.
Consider a small portion of the string between the points a and b, a < b. If p is the den-
sity of the string, the mass of this portion is p(b- a) and its acceleration is approximately
a2ujat 2 evaluated at some point x, a < x <b. So, by Newton's second law of motion the

force acting on this portion of the string is

.,
a-u
f = p(b- a)-., . (5.16)
at-
We can also calculate this force by another argument. This force arises from the tension r
of the string. We can assume r is constant when vibrations are small. This force acts along
the string. Its component in the y-direction is obtained by multiplying it by the sine of the
angle of inclination, and this is [1 + (au;ax) 2r 2au;ax.
1
' Therefore, if au;ax is much
smaller than I, as is the case when the vibrations are small, this is nearly equal to au;ax.
So the total force acting on the portion of the string between a. b is approximately

aulb .
f=T- (5.17)
ax a
Now equate (5.16) and (5.17), divide by b- a and let b-+ a. This gives
., .,
a-u T a-u
-=--.,.
at2 pax-
This is usually written as
.,
a-u .,
., a-u
-=a--.,,
at2 ax-
(5.18)

and called the one-dimensional wave equation.

Remark 5.3.1. Note that by a "dimension analysis" we can see what the factor a in (5.18)
2
represents. The units of r are those of force, i.e., mass x length x (time)- , the units of P
102 FOURIER SERIES

are those of density, i.e., mass x (length)- 1• So T 1p has the units of (length) 2 x (time)- 2,
i.e., of (velocity)2 . Thus a is a quantity like a velocity. This can be seen by analysing (5.18)
also. We will find this information useful, even though it does not affect any mathemati-
cal calculation. In fact, in several mathematics books you will find the statement '·a is a
constant, which may be put equal to I without loss of generality.''

In addition to (5.18) our system satisfies the boundlll)' conditions

u(O, t) = u(L. t) = 0 for all t. (5.19)

This says that the end points of the string are fixed throughout the motion. The initial
position and velocity can be chosen at will. This is stated as initial conditions

u(x. 0) = f(x). u 1 (x. 0) = g(x), (5.20)

where f and g describe the initial position and the initial velocity of a particle on the string
at the position x. Since we are assuming that u satisfies (5.18), both f and g must be C 2
functions (at least piecewise C2 ).
The problem is to find u that satisfies (5.18)- (5.20).
One way is to proceed as in Section 1.2 where we studied the heat problem. Assume
that there is a solution of the form

u(x. t) = X (x)T(t), (5.21)

where X is a function of x only and Tis a function t only. Then (5.18) leads to the equation

X" 1 T"
X- a2T

Since the left-hand side is independent oft, and the right-hand side is independent of x,
there is a constant c such that
X" I T"
-=.....,-=-c.
X a-T
This leads to two second order ordinary differential equations

X"+ eX= 0,
II "'l
T +a-cT= 0. (5.22)

Both equations arc of the type ( 1.6) encountered earlier.

When c = 0, the equations (5.22) lead to the solution

u(x. t) =(Ax+ B)(Ct +D),

where A. B. C, Dare constants. Now if C # 0. u(x, t) becomes unbounded as t ---? 00 .

We know from physical considerations that u must remain bounded. So C = 0. This means
that u is independent of r. So there cannot be any motion. This is contrary to what we know.
SOME APPLICATIONS 103

If c =ft 0. then proceeding as in Section 1.2 we get the solution

u(x. t) = (Aei.w + Be-i.ra)(Ceit/l + De-i1fl), (5.23)

where ex = ,JC. f3 = .JCa and A. B. C, D arc constants. If c > 0. then both ex. f3 are real
and we can assume without loss of generality that both are positive-otherwise we can
interchange A and B. and C and D. If c < 0, then both ex, f3 are imaginary.
From the first condition in (5.19). and (5.23) we get

0 = u<O. t) =(A+ B)(CeitfJ + De-i'fJ)

for all t. If the second factor on the right-hand side were zero, then u (x. t) would be iden-
tically zero. This is not possible. Hence

A =-B. (5.24)

Now using the second of the boundary conditions (5.19), (5.23), and (5.24) we get

(5.25)

If c < 0. then ex = iy for some real y =ft 0. and (5.25) becomes

which implies A = 0 and hence from (5.24) B = 0 as well. Again. this would mean that
u(x. t) is identically zero. and that is not possible.
So, we are left with the case c > 0. Now ex is real and positive, and (5.25) gives

A sin Lex= 0.

Since A cannot be zero (as we saw above) this condition implies that
llT!
ex=- /l = 1. 2.....
L
Then we must have
/lT{
f3 =-a.
L
Hence for each positive integer n we have a solution of (5.18) given by

intw
u 11 (x. t) =
• llT!X
sm - - (A"e + Bne -imw ), (5.26)
L
where A 11 , B11 are some (new) constants and w = rra/ L.
A finite linear sum of such u 11 would be a solution of (5.20) and so would be the infinite
sum
ex;
intu•
"'"""'
u(x, t) = L._.,sin L(Ane
llT!X
+ Bne -intu• ). (5.27)
n=l

provided this series is convergent and can be differentiated tenn by tenn.

104 FOURIER SERIES

Exercise 5.3.2. Show that the assumption u E C 2 implies that the series (5.27) converges
uniformly and can be differentiated term by tenn.
The arbitrary constants Au and B, occun-ing in (5.27) are determined from the initial
condition (5.20). These give

oo nrrx
f(x) = L<A, + B,) sin L'
11=1

oo nrrx
g(x) = Linw(A,- B,)sin r· (5.28)
II= I

Thus Au + B, and in w( A, - B,) are the coefficients in the Fourier sine series for f and g
respectively. So, they are detennined uniquely by f and g. Hence A, and B, are uniquely
determined by f and g.

Remark 5.3.3. Since the quantity a is a velocity. as observed earlier, the quantity w has
the units of velocity x (length)- I = (time)- 1• Thus w can be thought of as a frequency.
Our solution (5.27) involves positive integral multiples n ur of this frequency. In music w
is called thefimdamental and 2w, 3w, ... are called the overtones. Note that

So w increases with tension and decreases with the length and the density of the string.
This con-esponds well with our practical experience.
D'Alembert's original solution was different from (5.27). We can get it from (5.27) as
follows. Expand the exponentials occuring there into cosines and sines, then use familiar
trigonometric identities to get (5.27) into the form

u(x, t) = 2I """ . (nrr

~(A,+ B,) [ sm LX + ntw ) +sin (nrr
L - ntw
X )]

- 2I """.
~I (A, - L + n t w) - cos (nrrx
B,) [ cos (nrrx L - n t w)] .

Now, replace w by rr L/a and rearrange terms to get

u(x. t) =?I """
~ [ (A,+ B,)sm
. -(at+
nrr nrr
x)- i(A,- B,) cos -(at+ x) ]
- L L
I """ [ (A,+B,)sin
-2~ nrr nrr ] (5.29)
L(at-x)-i(A,-B,)cosT(at-x).

So, one can write

u(x, t) = v(at + x)- v(at - x) (5.30)

where v is defined by (5.29) and (5.30). Note that v is a function of one variable.
D'Alembe1t's solution was given in the form (5.30). It is not necessary to come to it via
the route followed above. A much simpler way is the following. Let v be any c2 function
SOME APPLICATIONS 105

of period 2L. Then if we define u(x. t) by (5.30), we can easily check that it satisfies the
wave equation (5.18). and also the boundary conditions (5.19). If u(x, t) is now to satisfy
the initial conditions (5.20). then we must have

f(x) = u(x,O) = v(x)- v(-x),

g(x) = u,(x, 0) = v'(x)- v'(-x).

These two conditions completely determine v; the first one determines the odd part of v
and the second determines its even pmt.
Notice that (5.30) can be interpreted as a superposition of two waves travelling in the
opposite direction with the same speed a. As a function of x, the graph of v(at + x) has
the same shape as that of v(x) but is translated by a distance at to the left. This can be
thought of as a wave travelling to the left with speed a. The point 0 on the string is fixed.
The function -v(at - x) represents the wave v(ar + x) reflected at the origin. and now
travelling to the right with speed a. The solution (5.30) is a superposition of these two
waves.

5.4 Band matrices

In this section we demonstrate an application of ideas from Fourier series to a problem in
linear algebra.
Let M (n) be the space of n x n complex matrices. A matrix A is also a linear operator
on the Euclidean space C". The noml of A is its norm as an operator, defined as

HAll= sup{IIAxll: x E C", llxll = 1}. (5.31)

An operator U is said to be unital)' if it satisfies any of the following equivalent conditions:

I. U preserves inner products. i.e., (Ux. Uy) = (x, y) forallx, y. E C".

2. U preserves norms, i.e., II Ux II = llx II for all x E C".
3. U is invertible, and (its Hermitian conjugate) U* = u-l: i.e .. U U* = U* U = /, the
identity operator.
4. For each orthonormal basis e1 ••.• , e, in C", the vectors Ue1, .... Ue, also form an
orthonormal basis.

It is easy to see that

IIU A VII =II All for all unitary U, V. (5.32)

Exercise 5.4.1. Let w = e2rrif", a primitive nth root of unity. For 0 ~ j < n - 1. let h
be then-vector with components

"- -w
f ].k-
I
.jii
1 "li.
. 0 < k < n- 1.
106 FOURIER SERIES

Thus

1
fo = r.:- (1 , 1, . . . , 1) ,
yll

1t= ( 1, W, W 2 ••.• , W 11-1) ,

fI =
yll

etc. Show that fo, /1, ... , J,,_, is an orthonormal basis for C11 • Show that the matrix F
whose jth column is the vector /j is a unitary matrix.

Exercise 5.4.2. The permutation matrix

0 0 I
I 0 0
R= 0 0

0 0 1 0

is unitary. Let F be the matrix of Exercise 5.4.1 and let U = F R F*. Show that U is the
diagonal unitary matrix with diagonal entries 1, w, ... , w11 -l, i.e.,

U . ( I , w, w 2, .... w11-1) .
= dmg (5.33)

Exercise 5.4.3. A few important facts of Linear Algebra are listed in this exercise. Prove
them.

(i) The Spectral Theorem says that every Hermitian operator (one for which A = A*)
can be diagonalised by a "unitary conjugation"; i.e., there exists a unitary U such that

U AU* = diag()..,, ... , A11 ).

The numbers Aj are the eigenvalues of A. So, for a Hermitian operator,

(ii) For any operator A we have IIA11 2 =II A* All.

(iii) The positive square roots of the eigenvalues of A are called the singular values of A.
They are enumerated in decreasing order as s 1(A) > s2(A) > ... > s11 (A). We have
II All = Sl (A).

Exercise 5.4.4. Calculate the norms of the 2 x 2 matrices

B=[~: l·
SOME APPLICATIONS 107

You will find that HAll = J2. and IIBII = ~(I+ ./5). In this example, B is obtained by
replacing one of the entries of A by zero. But II B II > II A 11.

The rest of this section is concerned with finding relationships between II Bll and II All
when B is obtained from A by replacing some of its entries by zeros.
Divide A into r 2 blocks in which the diagonal blocks are square mallices. not neces-
sarily of the same sizes. The block-diagonal matrix obtained by replacing the off-diagonal
blocks Aij. i =ft j, by zeros is called a pinching (an r-piuching) of A. We write this as

C(A) = diag(A II· .... A,.,.). (5.34)

Thus a 2-pinching is obtained by splitting A as [Aij], i. j = I. 2. and then putting

C(A) = [ A~ I A~21· (5.35)

Exercise 5.4.5.
(i) Let U be the block-diagonal matrix U = diag(l. -I) where the blocks have the same
sizes as in the decomposition (5.35). Show that the 2-pinching
I
C(A) =-(A+ U AU*>. (5.36)
2
Hence C(A) < II All.
(ii) Show that the r-pinching (5.34) can be obtained from A by successively applying
(r- 1) 2-pinchings. Thus C(A) < II All for every pinching.
(iii) Show that there exist r mutually orthogonal projections Pj inC", P1 + · · · + P,. = 1.
such that the pinching (5.34) can be written as
r
C(A) = LPjAPj. (5.37)
j=l

When r = n. the pinching (5.34) reduces to a diagonal matrix. We write this as 'D(A)
and call it the diagonal part of A. In this case the projections Pj in (5.37) are the projections
onto the !-dimensional spaces spanned by the basis vectors.

Exercise 5...1.6.
(i) Let U be the diagonal unitary matrix (5.33). Show that
I 11-l
V<A> =- LuiAu*1 • (5.38)
ll .
j=
0

(ii) Any r-pinching can be described as

I r-1
C<A> =- L vi AV*i (5.39)
. o
r j=

where V is a diagonal unitary matrix. What is V?

108 FOURIER SERIES

(iii) Usethisrepresentationtogiveanotherproofoftheinequality IIC(A)II < IIAII.

The expressions (5.37) and (5.39) display a pinching as a (noncommutative) m•eraging

operation or a com'ex combination of unitary conjugates of A.
For I < j < n- I, let 'Dj(A) be the matrix obtained from A by replacing all its entries
except those on the jth superdiagonal by zeros. (A superdiagonal is a diagonal parallel to
the main diagonal and above it.) Likewise, let V_j(A) be the matrix obtained by retaining
only the jth subdiagonal of A. In consonance with this notation we say Vo(A) = V(A).
Let Uo be the diagonal matrix

Uo = d .Ja0a( e iO , e2i0 , ... , eniO) .

The (r. s) entry of the matrix UoAU0 is then ei(r-s){la,.5 • Hence

Vk(A) = ?I ITT eikO UoAU0* dO. (5.40)

_T( -TT
When k = 0. this gives one more representation of the main diagonal V(A) as an average
over unitary conjugates of A. From this we get the inequality II'Dk{A)II < IIAII. Using
(5.40) we see that

Let 1j(A) = V_, (A)+ Vo(A) + V, (A). This is the matrix obtained from A by keeping
its middle three diagonals and replacing the other entries by zeros. Again. using (5.40) we
see that

1173(A)II <
I
rr
ITT
-TT II +2cosO!dOIIAII.
2

Exercise 5.4.7. Calculate the last two integrals. You will get

(5.41)

(5.42)

Note that using the inequality II'Dk(A)II < IIAII and the triangle inequality we get the
bounds 211AII and 311AII. respectively, for the left-hand sides of (5.41) and (5.42). These
bounds are weaker than what we have obtained from the integrals.
A trimming of A is a matrix of the form

k
12k+l (A) = L 'Dj(A).
i=-k
 SOME APPLICATIONS
109
obtained by replacing all diagonals of A outside the band -k < j < k by zeros. We have
1'(

12k+l (A) =
1
-rr Dk(fJ)UuAU0dO.

where Dk(O) is the Dirichlet kernel (2.6). Hence. we have

(5.43)

where Lk is the Lebesgue constant. Since Lk = 0 (log k) this inequality is much stronger
than the inequality II'T:?k+I(A)II < (2k + l)IIAII that one gets by using just the triangle
inequality and II'Dj(A)II < II All.
Let !J.u (A) be the matrix obtained from A by replacing all entries of A below the main
diagonal by zeros. Then D.u is called the triangular truncation operat01:

~xercise5.4.8.

(i) Let B beak x k matrix and let

A= [ 0
B
B*
0
J.
This is a matrix of order 2k. Show that II A II = II B II.
(ii) Show that

(iii) Thus. if D.u is the triangular truncation operator on M(k). then

(5.44)

where Lk+l is the Lebesgue constant.

Let A be the k x k matrix with entries aij = ( i - j)- 1 if i =ft j. and a;;= 0. It
can be seen (with some labour) that HAll < rr. while IID.u(A)II = O(logk). Thus the
inequality (5.44) is nearly the best possible. This is an important fact. It corresponds to a
basic fact about Fourier series proved by M. Riesz. The analytic part of a Fourier series
L~-oo a,ei"o is the truncated series L~o a,ei"e. Riesz showed that the truncation map
that sends the Fourier series of a function to its analytic part is a bounded map on the
space L P if I < p < oo, but not if p = I or oo.
The matrix F of Exercise 5.4.1 is called the matrix of the Finite Fourier Transf01m. a
concept intimately related to Fourier series.

Recommended supplementary reading for this section: M.-D. Choi. Tricks or treats with
the Hilbert matrix, American Mathematical Monthly, Volume 90. May 1983. pp. 301-
312; and R. Bhatia, Pinching, trimming. truncating and averaging of matrices, American
Mathematical Monthly, Volume 107, August-September 2000. pp. 602-608.
 . . ...
' ·,. .•"
~ .1> I ). ' ~-.

A Note on Normalisation

One peculiar feature of writings on Fourier analysis is that the factor 2rr is placed at differ-
ent places by different authors. While no essential feature of the theory changes because of
it, this is often confusing. You should note that we have used the following notations:

1. The Fourier coefficients off are defined as

f(n) = -I iTT . ()
f(O)e- 111 dO.
2rr -rr

2. The convolution of f and g is defined as

(f * g)(x) = ~~ f(x- t)g(t) dt.

3. The Dirichlet kernel is defined as

I N
DN(t) = -2rr L
11=-N
i"'

4. For functions on a bounded interval/ we have defined L2 and L 1 norms as

llfll2 = ( -
Ill
1 1I
lf(x)l 2dx
) 1/2
,

1
II/III = - { lf(x)l dx,
Ill J1
where Ill is the length of the interval /.
When you read other books you should check whether the same conventions are being
followed. For example. some authors may not put in the factor I /Ill while defining norms~
others may put in the factor 1/2rr while defining f *g. Then some of their theorems will
111
112 FOURIER SERIES

look different because the factor 2rr if suppressed at one place will appear elsewhere. See
the remark following Theorem 4.2.4.
As a consolation we might note that in spite of several international conferences on
.. units and nomenclatures" many engineering texts usc-sometimes on the same page-
the f.p.s .• the c.g.s., the m.k.s. and the rationalised m.k.s. systems of units.
A Brief Bibliography

This is a brief bibliography. Some of the books listed here provide the necessary back-
ground. Others are suggested as collateral and further reading.

Analysis
Several generations of mathematics students have learnt their analysis from the classic

W. Rudin, Principles of Mathematical Analysis, McGraw Hill. first published in

1953, 3rd ed., 1976.

Chapters 1-8 of Rudin provide adequate preparation for reading most of this book. For
some of the sections more advanced facts about integration are needed. These may be
found in Chapter I 0 of Rudin. and in greater detail in Part I of another famous text:

H. L. Royden. Real Analysis, MacMillan. 3rd ed .. 1988.

Elementary facts about differential equations that we have used in this book are gener-
ally taught as applications of the Calculus. These may be found. for example. in

R. Courant, Differential and Integral Calculus, 2 volumes. Wiley Classics Library

Edition. 1988.

or, m

W. Kaplan, Ad,·anced Calculus, Pearson Addison-Wesley. 5th ed .. 2002.

113
114 FOURIER SERIES

For a more detailed study of differential equations the reader may see

W. E. Boyce and R. C. DiPrima. Elementary Differential Equations and Bowulmy

Value Problems. 7th ed .. Wiley Text Books, 2002.

Chapter I 0 of this book, titled "Partial differential equations and Fourier Series." con-
tains topics close to the ones we have studied in the early sections.
We have used elementary properties of complex numbers and functions in this book.
At places we have pointed out connections with deeper facts in Complex Analysis. The
standard reference for this is

L. V. Ahlfors. Complex Analysis. 3rd ed., McGraw Hill. 1978.

Fourier series
Two well-known books on Fourier series, available in inexpensive editions. are

H.S. Carslaw, Introduction to the theory of Fourier's Series and Integrals, Dover
Publications, 1952.

and

G.H. Hardy and W.W. Rogosinski. Fourier Series, Cambridge University Press,
1944.

Less well known and, unfortunately, not easily available is the excellent little book

R.T. Seeley. An Introduction to Fourier Series and Integrals, W.A. Benjamin Co.,
1966.

Our beginning sections are greatly influenced by, and closely follow. Seeley's approach. A
more recent book which is very pleasant reading is

TW. Korner. Fourier Analysis, Cambridge University Press, paperback edition.

1989.

All these books are at an intem1ediate level. More advanced books, for which a thorough
knowledge of functional analysis is essential include

H. Helson, Harmonic Analysis, Hindustan Book Agency, 1995,

Y. Katznelson, An Introduction to Harmonic Analysis. Dover Publications. 1976.

and

H. Dym and H.P. McKean. Fourier Series and Integrals, Academic Press. 1972.
ABRIEF BIBLIOGRAPHY 115

General reading
We have emphasized the history of Fourier Series and their connections with other subjects.
The following books were mentioned in the text.

W. Dunham, Joumey Through Genius, Penguin Books, 1991.

W. Dunham, Euler: The Master of Us All, Mathematical Association of America,

1999.

V. M. Tikhomirov, Stories about Ma.-..:ima and Minima, American Mathematical So-

ciety, 1990.

E. Maor, Trigonometric Delights, Princeton University Press, 1998.

J. Arndt and C. Haenel, rr Unleashed, Springer, 200 I.

A. Weil, Number Theory: An Approach Through Histmy from Hammurapi to Legen-

dre, Birkhauser, 1984.

History and biography

The recent book

J.-P. Kahane and P.-G. Lemarie-Rieuseet, Fourier Series and Wavelets, Gordon and
Breach, 1995

provides a scholarly history of Fourier Analysis and also a treatment of more recent topics.
Fourier's original book Theorie Ana/ytique de Ia Chaleur has been translated into En-
glish as The Analytical Tlzemy of Heat published in a paperback edition by Dover. It is
regarded as both a scientific and a literary masterpiece. A biography of Fourier is con-
tained in the famous book Men of Mathematics by E.T. Bell, Simon and Schuster, 1937. A
scholarly full length biography is Joseph Fourie1; The Man and the Physicist by J. Heriv-
ell, Oxford, 1975. A few facts about Fourier may induce you to look at these books. He
was a revolutionary, accused of being a "terrorist" and was almost killed on that charge;
he was a soldier in Napoleon's army that invaded Egypt and later a successful governor of
a province in France: he wrote a survey of Egyptian history that has been acknowledged
as an outstanding work by historians; he was the Director of the Statistical Bureau of the
Seine and is known among demographers for his role in developing governmental statistics
in France; his book on heat propagation has been repeatedly published as a part of "Great
Books" collections by several publishers. A most striking counterexample to the common
belief that a scientist is someone forever bent over his equipment or papers, oblivious of
and indifferent to the world around.
116 FOURIER SERIES

... if man wishes to know the aspect of the heal'-

ens at successil'e epochs separated by a great
number of centuries, if the actions of gravity and
of heat are exerted in the interior of the earth at
depths which will always be inaccessible. math-
ematical analysis call yet lay hold of the laws of
these phenomena. It makes them present and mea-
sw·able, and seems to be a faculty of the human
mimi destined to supplement the shortness of life
and the impelfection of the senses; and what is
stillmore remarkable, it follows the same course
in the stlldy of all phenomena; it illle1prets them
by the same language, as if to attest the unity and
simplicity of the plan of the universe, and to make
stillmore el'ident that unchangeable order which
presides over all natural causes.
Joseph Fourier
Index

Abel limit. 28 ergodic principle, 95

Abel summability. 28 Euler constant. 33
algebra. 89 exponcnti<tl polynomiuls. 33
approximate identity. 21
Fejer kernel. 31
Banach algebra. 89 Fejer's theorem. 31
Banach space, 48 Fourier coefficients
band matrices, I 05 of L 1 functions. 85
Basel problem. 75 of L2 functions, 81
Bernoulli numbers. 65 uniqueness of. 24
Bessel's inequality. 81 with respect to an orthonornutl system. 81
boundary value problem, 15 convolution. 88
order of magnitude of. ...J.O
Cauchy-Schwarz inequality. 81 rate of decay, 91
Ces~trosummability. 28 rel<ttion with smoothness. 43
convolution. 20. 85 Fourier series. 19. 28, 31
and Fourier coefficients. 89 L 2 convergence of. 79
and smoothness, -14, 90 <tnd closest approximation. 82
lack of an indentity for, 90 examples. 55
cotangent. 62 cosine series. 51, 52
definition. 19
Dido's problem. 100 divergence of. 46
Dirac family, 22 exponential fonn. 52
Dirac sequence, 21 pointwise convergence of, 35
Dirichlet kernel. 29 tennwise integration. 45
Dirichlet problem, 15 trigonometric fonn. 52
solution. 23 Fubini 's Theorem. 86
uniqueness of solution, 24 function
Dirichlet's theorem, 43 on T,l9

117
118 FOURIER SERIES

function (cominued) Poisson integral, 20

absolutely continuous, 44 Poisson kernel, 20
bounded variation, 41 Poisson's theorem, 23
even and odd, 51 positive definite sequence, 73
integrable, 26 positive operator, 74
Lipschitz continuous, 38 principle of localisation, 38
periodic, 19 pulse function, 55
piecewise C 1, 36
rapidly decreasing, 84
Gibbs phenomenon, 70 Riemann-Lebesgue Lemma, 36, 88
Riesz-Fischer theorem, 83
hannonic function, 19
Hausdorff moment theorem, 53 saw-tooth curve, 57
Heaviside function, 57 separation method, 17
Hilbert space, 79 steady flow of heat, 13

infinite products, 61 Tchebychev polynomial, 53

integrable function, 25 temperature
isopcrimetric problem, 98 maximum and minimum. 25
mean value property, 25
Jordan's theorem, 43 Tonelli's theorem, 86
triangular truncation operator. I09
Korovkin's theorem, 74 triangular wave, 57

Laplace equation unifonn boundedness principle, 48

solution of, 19 unitary matrix, I05
Lebesgue constants. 33
Legendre polynomials, 83 vibrating string, I0 I

Newton ·s law of cooling, 13 Wallis formula. 65

Newton's Second Law, I0 I wave equation. I0 I
nonn, 105 Weierstrass approximation theorem, 25, 53
Weyl's equidistribution theorem, 97
Parseval's relations, 83 Weyl's theorem. 96
pennutation matrix, I06 Wirtinger's inequality, 85
pinching. I07
Phmcherel's theorem, 82 zeta function, 67
 Index
119
Notation
c 1,25
c 2, 13
D. 15
DN(t), 29
F,, (t), 31
L, (!), 80
L2(/), 80
L,,33
0(*). 40
P(r, 8), 20
Pr, 23
SN(f; 8), 27
T, 19
Si(x), 67
a,,28
s(s), 67
f * g, 20
/2.79
o(*), 44
SN, 28
j(n), 19
120 FOURIER SERIES

About the Author

Rajendra Bhatia received his PhD from the Indian Statistical Institute in New Delhi,
India. He has held visiting appointments at various universities across the world-Sapporo,
Kuwait, Ljubljana, Pisa, Bielefeld, Lisbon, Toronto. and Berkeley. among others.

He is a member of the Indian Mathematical Society, the MAA, the AMS, and the Inter-
national Linear Algebra Society (on whose board of directors he has served). He has been
on the editorial boards of several mathematics journals such as Linear Algebra and Its Ap-
plications, S/Aiv! Journal on Matrix Analysis, and Linear and Multilinear Algebra. He has
been a recipient of the Indian National Science Academy Medal tor Young Scientists and
Shanti Swarup Bhatnagar Prize of the Indian Council for Scientific and Industrial Research.
He is a Fellow of the Indian Academy of Sciences.

His previous widely acclaimed books include: Perturbation Bounds for Matrix Eigenval-
ues, and Matrix Analysis.
foURIER JERIES
11Jis book is ,1 ·i't'l~l' rc.',Jthlblc introduc:tion tlJ Fouric.•r Si'ries ruit,Jb!t·for scientists
,111d c.·m.:.int'c.'rs. It is sprinklc.·d ·u.·itb bints ,Jbmtt more.· r.:ant dc.''i.'c.·lopmc.'llts ,u1d b,Js
,/ lot t~(nia bistoric.",J! c."fl/1/lllc.'nt, tb,Jt 'U.•i/1 intr~e,uc tbc be_,f studc.'llfS ,md I!Mtb
m,~jorJ. 11,e .mtblJr ,JIIIIfl.\1 ltilks to tbt~ re,tdcrJ ,md Jkil!ful~r b1:e.bl~e.btJ u.!b.Jt
t~f impor/,tllt. A _t:1ir ,//llfllllll of tbt• llt.lft'ri,J! j_, in tbe t'Xt.:nsh:c set of
t•.wrdJc.'J. ~{tbiJ 'i'el:) m(:c text bt~d been ,/'i.'tliltb!t• u.'bc.·n I ·u.•,JJ lt'tJ(bil~f!.- I 'L..'ottld
b,J'i.'t' ust·d it f or .1 junior-.rc.·nior lc.''i't'l amnc.•for sdc.·nce t111d llltilb 111t1jon.

~ hc..·nnc..·lh .\. Ro~~.

l nt\l· r,rl~ or Ort·gon. Lugc..·ne

'1:.' " .1 " " " 1w mil no luo """ In I uu 11<1 '<II<., < " ' c 1111)! lu ''""· lll.ll"' llw mo·,.
rhc..·urc..·m'. c..· \..nnplt-, .m.l .rpphl.lllnn' h l.lll ltl· u'c..·cl h• k.r r n lite..· 'Uitlcll. .mel .tJ,.. ln
'liJlJ ,Junllll l·nh.lllll .m•lc.. mi.dJ.,h llll•lt· r~r.ulu.lll' l• •ur,c..-, • •n 111.11 hu11.11u .11 .111.1h ' ''

lin J,. ,.,J.. I.t-~111, \\ ll h .1 I.ud 'umm.u' nl rlll" fit h hr'h ••' n t I nUIIl-r ' l"fll"' .,, l'f rllfn.·
• t"lllllfll'' I he 'lll•tn r '' p~t:,t· mc..-.t Ill .1 \\,1\ rh.ll • n.rl•ll-, dtc..· fl·.ult·r In .1pprl·ll.lll" hem
.1 111.11 hun.rr u .rl rlwul\ • lt·H·Iup' Ill ''-';!l"' I rum .1 pr.u l u .II prultkm hue h ·'' lonclul unn
•• I hc..·.rll '" .111 .rJ.,rr.lll rhcnl\ .ll .rl nt_!.! \\llh ennct·p•' 'll• h ,1, ,c..·h . hlllllhlll'. rnllllU\ .mel
e oil\ e f!!l ne e I ht .11"' 1.1• I rlwor \ l ht 11 J'l ll\ltle·' 1111 lore 'lTll .rpplu .II loll' Ill clr\ l-hl' . IlL I'

llu .n ulaor , 1,111' nul \\llh .1 cle" 11p11on nl r ht prultluu 1h.11 lnl l nllllll rn llllln•hltl Ju,
l.1111o11' " Ill'' llu· m.rrhnu.llu .rl Jllnltluu' rJu, lc .r.J, Itt .Ill lllt'Jl elr''""n l ll~.!.t'lttll'l'
I \..uupll '· ,.,,., c '""' .mel .In e • 111111, It •• hu r he 1 11 .rcl m.c .llh I ll·" .Ill h .11 e pro\ rclt •I. .rlnn~ \\ ll h
.1 e h.apr• 1 rh.u P'"''.ll·, 111.11• u.rl.u .1 mn r• .uh.uu c .lltHI 'llll.lhlt lnr ~r.rclu. n t· 'llldllll'
I lu .mrl1n1 de liii•II' II.JI• ' .rpph• .1111 llh nl rlw rlu·nr \ It • .1 J,,.,,,,J l .lll;_!c nl p rnltlun'

lin c '' •••" ."'' \.11\lll;_!ln•l,ot drllh ull\ rlt.ar .lie" .ll h rnl rh ruu;_:hnul r ht J,.,.,J.. \\ Ill hd p
•• ·.11J. ' ' 11 ' ' du 11 1111tlc ,,l.llldlll.t:. nl rlu lll.ll l ll.ll

•
1111
/HII BH"\
Ill
R 57 '1 11~