Syllabus Fourier analysis

by T. H. Koornwinder, 1996

Universiteit van Amsterdam, Faculteit WINS, vakgroep Wiskunde

Last modi
ed: November 1998
Note This syllabus is based on parts of the book \Fouriertheorie" by A. van Rooij (Epsilon
Uitgaven, 1988). Many of the exercises and some parts of the text are quite literally taken
from this book. Usage of this book in addition to the syllabus is recommended. The
present version of the syllabus is slightly modi
ed. Modi
cations were made by J.J.O.O.
Wiegerinck.
Contents
Part I. Fourier series
1. L2 theory.
2. L1 theory
3. The Dirichlet kernel
4. The Fejer kernel
5. Some applications of Fourier series
Part II. Fourier integrals
6. Generalities
7. Inversion formula
8. L2 theory
9. Poisson summation formula
10. Some applications of Fourier integrals
Standard references
1] A. van Rooij, Fouriertheorie, Epsilon Uitgaven, 1988.
2] A. A. Balkema, Syllabus Integratietheorie, UvA, FWI, 1993.
3] J. Wiegerinck, Syllabus Functionaalanalyse, UvA, FWI, 1994.
4] W. Rudin, Real and complex analysis, McGraw-Hill, Second edition, 1974.
5] T. W. Korner, Fourier analysis, Cambridge University Press, 1988.
Further references
6] H. Dym & H. P. McKean, Fourier series and integrals, Academic Press, 1972.
7] Y. Katznelson, An introduction to harmonic analysis, Dover, Second edition, 1976.
8] J. Korevaar, Syllabus Fourier Analysis, UvA, FWI, 1991.
9] T. W. Korner, Exercises for Fourier analysis, Cambridge University Press, 1993.
10] W. Schempp & B. Dreseler, Einfuhrung in die harmonische Analyse, Teubner, 1980.
11] E. M. Stein & G. Weiss, Introduction to Fourier analysis on Euclidean spaces, Princeton University Press, 1971.
12] G. Weiss, Harmonic analysis, in Studies in real and complex analysis, I. I. Hirschman
(ed.), MAA Studies in Math. 3, Prentice-Hall, 1965, pp. 124{178.
13] A. Zygmund, Trigonometric series, Cambridge University Press, Second ed., 1959.

1 L2 theory
1.1 The Hilbert space L22
1.1 T -periodic functions. Let T > 0. A function f : R ! C is called periodic with period

T (or T -periodic) if f (t + T ) = f (t) for all t in R . A T -periodic function is completely

determined by its restriction to some interval a
a + T ), and any function on a
a + T ) can
be uniquely extended to a T -periodic function on R . However, a function f on a
a + T ] is
the restriction of a T -periodic function i f (a + T ) = f (a). When working with T -periodic
functions we will usually take T := 2.

1.2 The Banach spaces Lp2 . Let 1
p < 1. We denote byRLp2 the space of all 2periodic (Lebesgue) measurable functions f : R ! C such that ; jf (t)jp dt < 1. This

is a complex linear space. In particular, we are interested in the cases p = 1 and p = 2.

The reader may continue reading with these two cases in mind. A seminorm k : kp can be
de
ned on Lp2 by

1=p
Z
1
p
kf kp := 2 jf (t)j dt
(f 2 Lp2 ):
(1:1)
;
By seminorm we mean that kf kp  0, kf + gkp
kf kp + kgkp and k f kp = jj kf kp for
f
g 2 Lp2 ,  2 C , but not necessarily kf kp > 0 if f 6= 0. The factor (2);1 is included in
(1.1) just for cosmetic reasons: if f is identically 1 then kf kp = 1.
It is known from integration theory (see syll. Integratietheorie, Chapter 6) that, for
f 2 Lp2 , we have that kf kp = 0 i f = 0 a.e. (almost everywhere), i.e., i fp(t) = 0 for t
outside some subset of R of (Lebesgue) measure 0. It follows that, for f 2 L2 , the value
of kf kp does not change if we modify f on a set of measure 0, so k : kp is well-de
ned
on equivalence classes of functions, where equivalence of two functions means that they
are equal almost everywhere. Also, kf kp = 0 i f is equivalent to the function which is
identically 0 on R .
Thus we de
ne the space Lp2 as the set of equivalence classes of a.e. equal functions
p
in L2 . It can also be viewed as the quotient space Lp2 =ff 2 Lp2 j kf kp = 0g. The space
Lp2 becomes a normed vector space with norm k : kp . In fact this normed vector space is
complete (see syll. Integratietheorie, Chapter 6 for p = 1 or 2), so it is a Banach space.
If I is someRinterval then we can also consider the space Lp (I ) of measurable functions
on I for which I jf (t)jp dt < 1, and the space Lp (I ) of equivalence classes of a.e. equal
functions in Lp (I ). A seminorm or norm k : kp on Lp (I ) or Lp (I ), respectively, is de
ned
by

1=p
Z
p
kf kp := jf (t)j dt :
(1:2)
: Lp

The linear map f 7! f j;  ) 2 ! Lp (;
)) is bijective and it preserves k : kp up to

a constant factor. It naturally yields an isomorphism of Banach spaces (up to a constant
factor) between Lp2 and Lp (;
]). Note that, when we work with Lp2 rather than Lp2 ,
the function values on the endpoints of the interval ;
] do not matter, since the two
endpoints form a set of measure zero.

L2 THEORY

Ex. 1.3 Show the following. If f 2 L12 then

Z

f (t) dt =

+a

; +a

(a 2 R ):

f (t) dt

1.4 The Banach space C2 . De
ne the linear space C2 as the space of all 2-periodic
continuous functions f : R ! C . The restriction map f 7! f j;  ] identi
es the linear space
C2 with the linear space of all continuous functions f on ;
] for which f (;) = f ().
The space C2 becomes a Banach space with respect to the sup norm

kf k1 := sup jf (t)j = sup jf (t)j

t2R

t2;  ]

(f 2 C2 ):

We can consider the space C (;
]) of continuous functions on ;
] as a linear
subspace of Lp(;
]), but also as a linear subspace of Lp (;
]).
Indeed, if two continuous functions on ;
] are equal a.e. then they are equal everywhere.
Hence each equivalence class in Lp (;
]) contains at most one continuous function.
Hence, if f
g 2 C (;
]) are equal as elements of Lp (;
]) then they are equal as
elements of C (;
]).
It follows that we can consider the space C2 as a linear subspace of the space Lp2 , but
also as a linear subspace of Lp2 .
The following proposition is well known (see Rudin, Theorem 3.14):

1.5 Proposition C (;
]) is a dense linear subspace of the Banach space Lp (;
]).
Ex. 1.6 For p  1 prove the norm inequality

kf kp
kf k1

(f 2 C2 ):

Ex.
1.7 Prove that the linear space C2 is a dense linear subspace of the Banach space
Lp2 (p  1).
Hint Show for f 2 C (;
]) and " > 0 that there exists g 2 C (;
]) such that
g(;) = 0 = g() and kf ; gkp < ".
Ex. 1.8 Prove the following:

Let V be a dense linear subspace of C2 with respect to the norm k : k1 . Then, for p  1,
V is a dense linear subspace of Lp2 with respect to the norm k : kp .

1.9 For f a function on R and for a 2 R de
ne the function Ta f by

(Ta f )(x) := f (x + a)

(x 2 R):

(1:3)

If f is 2-periodic then so is Ta f and, if V is any of the spaces Lp2 or C2 then Ta : V ! V

is a linear bijection which preserves the appropriate norm.

CHAPTER 1

Proposition Let p  1, f 2 Lp2 . Then the map a 7! Ta f : R ! Lp2 is uniformly

continuous.
Proof First take g 2 C2 . Then, by the compactness of ;
] and by periodicity,
g is uniformly continuous on R . Hence, for eachp " > 0 there exists  > 0 such that
kTpa g ; Tb gk1 < " if ja ; bj < . Next take f 2 L2 and let " > 0. Since C2 is dense in
L2 (see Exercise 1.7), we can
nd g 2 C2 such that kf ; gkp
31 ". Hence

kTa f ; Tb f kp
kTa (f ; g)kp + kTa g ; Tb gkp + kTb (f ; g)kp
23 " + kTa g ; Tb gk1

where we used Exercise 1.6. Now we can
nd  > 0 such that kTa g ; Tb gk1 < 31 " if
ja ; b j <  .

1.10 The Hilbert space L22 . We can say more about Lp2 if p = 2. Then, for any two
functions f
g 2 L22 , we can de
ne hf
gi 2 C by
Z
1
hf
gi := 2 f (t) g(t) dt:
;

(1:4)

Note that hf
f i = (kf k2)2 . The form h :
: i has all the properties of a hermitian
inner product on the complex linear space L22 , except that it is not positive de
nite, only
positive semide
nite: we may have hf
f i = 0 while f is not identically zero. However, the
right hand side of (1.4) does not change when we modify f and g on subsets of measure
zero. Therefore, hf
gi is well-de
ned on L22 and it is a hermitian inner product there. In
fact, the inner product space L22 is complete: it is a Hilbert space.
For I an interval and f
g in L2(I ) or L2(I ) we can de
ne hf
gi in a similar way:
Z
hf
gi := f (t) g(t) dt:
(1:5)
I

The bijective linear map f 7! f j;  ): L22 ! L2 (;
)) preserves h :
: i up to a

constant factor. It naturally yields an isomorphism of Hilbert spaces (up to a constant
factor) between L22 and L2 (;
]).
The space L22 is a linear subspace of L12 . In fact, for f 2 L22 we have the norm
inequality
kf k1
kf k2:
(1:6)
(Prove this by use of the Cauchy-Schwarz inequality.) Also, L22 is a dense linear subspace
of L12 . Indeed, the subspace C2 of L22 is already dense in L12 .

1.11 An orthonormal basis of L22 . The functions t 7! eint (n 2 Z) belong to C2 and

they form an orthonormal system in L22 :
1 Z eimt eint dt = 
mn
2 ;

(m
n 2 Z):

(1:7)

By a trigonometric polynomial on R (of period 2) we mean a
nite linear combination

(with complex coecients) of functions t 7! eint (n 2 Z).
The following theorem has been mentioned without proof in syll. Functionaalanalyse.

L2 THEORY

Theorem

(a) The space of trigonometric polynomials is dense in C2 with respect to the norm k : k1 .
(b) The space of trigonometric polynomials is dense in L22 with respect to the norm k : k2.
(c) The functions t 7! eint (n 2 Z) form an orthonormal basis of L22 .
Part (b) of the Theorem follows from part (a) (why?), and part (c) follows from part
(b). We will prove the theorem in a later chapter (
rst part (b) and hence part (c), and
afterwards part (a)). However, in this Chapter we will already use the Theorem. So we
have to be careful later that circular arguments will be avoided.

1.2 Generalities about orthonormal bases

1.12 We now recapitulate some generalities concerning orthonormal bases of Hilbert
spaces, as given in syll. Functionaalanalyse. Let H be a Hilbert space. Denote the inner
product by h :
: i and the norm by k : k. Let A be an index set and consider an
orthonormal system E := fe g2A in H, i.e., he
e i =  for

2 A. For convenience
we assume that the Hilbert space is separable. Thus the index set A of the orthonormal

system E will be countable.

Proposition (Bessel inequality)

X

2A

jhf
eij2
kf k2

(f 2 H ):

Corollary If f 2 H then lim!1 hf
ei = 0, i.e., for all " > 0 there is a
nite subset
B  A such that, if 2 AnB then jhf
eij < ".
(The set A, equipped with the discrete topology, is locally compact. By adding the point
1, we obtain the one-point compacti
cation of A. This gives a further explanation of the
notion lim!1 .)

Denition-Theorem (orthonormal basis Parseval's norm equality)

The orthonormal system E is called an orthonormal basis of H if the following equivalent

properties hold:
(a) P
Span(E ) is dense in H.
(b) 2A jhf
eij2 = kf k2 for all f 2 H (Parseval equality).
(c) If f 2 H and f is orthogonal to E then f = 0 (i.e., E is a maximal orthonormal
system).

1.13 Denition-Proposition (unconditional convergence)

Let H be a separable Hilbert
space. Let A be a countably in
nite index set. Let fv g2A 
H. We say that the sum P2A v unconditionally converges to some v 2 H if the following

two equivalent properties are valid:

P
(a) For each way of ordering A as a sequence 1
2
: : : we have that v = limN !1 Nn=1 vn
in the topology of H.
(b) For each " > 0 there is a
niteP
subset B  A such that for each
nite set C satisfying
B  C  A we have that kv ; 2C v k < ".

CHAPTER 1

Theorem Let H be a separable Hilbert spacePwith orthonormal basis fe g2A (A a

countable index set). Let fc g2A 2 `2(A) (i.e. 2A jcj2 < 1). Then thereP
is a unique
f 2 H such that hf
ei = c ( 2 A). This element f can be written as f = 2A c e
(with unconditional convergence).

Proposition (Parseval's inner product equality)

Let H be a separable Hilbert space with orthonormal basis fe g2A (A a countable index
set). Then
X
hf
gi = hf
ei hg
ei (f
g 2 H)
with absolute convergence.

2A

1.14 Theorem (summarizing this subchapter) Let H be a separable Hilbert space

with orthonormal basis fe g2A (A a countable index set). Then there is an isometry of
Hilbert spaces

F : f 7! fc g2A: H ! `2(A)

de
ned by c := hf
ei, with inverse isometry

F ;1: fcg2A 7! f : `2(A) ! H

P
given by f := 2A c e (unconditionally).

1.3 Application to an orthonormal basis of L22

By Theorem 1.11 the functions t 7! eint (n 2 Z) form an orthonormal basis of L22 . Let us

apply the general results of the previous subsection to this particular orthonormal basis of
the Hilbert space L22 .

1.15 Denition Let f 2 L12 . The Fourier coecients of f are given by the numbers
Z
1
fb(n) := 2
f (t) e;int dt
;

(n 2 Z):

(1:8)

Note that the integral is absolutely convergent since jf (t) e;int j
jf (t)j.
We can consider fb as a function fb: Z ! C . The map F : f 7! fb, sending 2-periodic
functions on R to functions on Z, is called the Fourier transform (for periodic functions).
Thus, by the Fourier transform of a function f 2 L12 we mean the function fb.

Remark In this subchapter we will consider the Fourier transform f 7! fb restricted to

functions f 2 L22 . For such f we can consider the right hand side of (1.8) as an inner
product. Indeed, if we put en (t) := eint (n 2 Z) then fb(n) = hf
eni (f 2 L22 ).
Proposition (Bessel inequality)
Z
1
X
1
2
b
jf (n)j
2 jf (t)j2 dt

n=;1

(f 2 L22 ):

Corollary (Riemann-Lebesgue Lemma) If f 2 L22 then limjnj!1 fb(n) = 0.

L2 THEORY

1.16 The results of x1.15 did not depend on the completeness of the orthonormal system.
However, the proof of the next results uses this completeness, so we cannot be sure about
these results until we have proved Theorem 1.11(b).

Theorem (Parseval) Let f
g 2 L22 . Then

1
1 Z jf (t)j2 dt = X
b(n)j2

j
f
2 ;
n=;1
Z
1
X
1
b
2 ; f (t) g(t) dt = n=;1 f (n) gb(n):

(1:9)
(1:10)

In (1.9) and (1.10), the integral on the left hand side and the sum on the right hand side
are absolutely convergent.
1.17 Theorem (Riesz-Fischer) Let fngn2Z 2 `2 (Z), i.e. P1n=;1 jnj2 < 1. Then
there is a unique f 2 L22 such that fb(n) = n (n 2 Z), and this f can be written as
1
X
n einx (unconditional convergence in L22 ):
f (x) =
So, in particular,

n=;1

f (x) = Nlim
!1
i.e.,

N
X
n=;N

n einx

(convergence in L22 ),

Z 
N
2
X

int
lim
f
(
t
)
;

e

 dt = 0:
n
N !1
;

n=;N

1.18 (summarizing this subchapter) The map F : f 7! fngn2Z: L22 ! `2 (Z) given by

Z
1
n := 2
f (t) e;int dt
;
de
nes an isometry of Hilbert spaces with inverse isometry F ;1: fngn2Z 7! f : `2(Z) ! L22
given by
1
X
f (x) :=
n einx (unconditional convergence in L22 ):
n=;1

1.19 Without proof we mention a more recent, very important and deep result: the
almost everywhere convergence of Fourier series of any function in L22 .

Theorem (L. Carleson, Acta Mathematica 116 (1966), 135{157) If f 2 L22 then
N
X
f (x) = Nlim
fb(n) einx
!1
n=;N

with pointwise convergence for almost all x 2 R .

CHAPTER 1

1.4 Orthonormal bases with cosines and sines

The orthonormal basis of functions t 7! eint (n 2 Z) for L22 (see Theorem 1.11(c))
2

immediately gives rise to an orthonormal basis for L2 in terms of cosines and sines. We
formulate it in two steps and leave the straightforward proofs to the reader.
1.20 Lemma For n = 1
2
: : : the two-dimensional subspace of L22 spanned by the
int also has an orthonormal basis given by the two
two orthonormal
p functions t 7! ep
functions t 7! 2 cos(nt) and t 7! 2 sin(nt).
p
p
Proposition The functions t 7! 1, t 7! 2 cos(nt) (n = 1
2
: : :), and t 7! 2 sin(nt)
(n = 1
2
: : :) form an orthonormal basis for L22 .
Note that, in the orthonormal basis of the above Proposition, the cosine functions
(including 1) are even and the sine functions are odd. In fact, we can split the Hilbert
space L22 as a direct sum of the subspace of even functions and the subspace of odd
functions, such that the cosines form an orthonormal basis for the
rst subspace and the
sines an orthonormal subspace for the second subspace. Let us
rst discuss the notion of
direct sum decomposition.
1.21 Denition Let H be a Hilbert space and let H1 and H2 be closed linear subspaces
of H (so H1 and H2 are Hilbert spaces themselves). We say that H is the (orthogonal)
direct sum of H1 and H2 (notation H = H1 H2 ) if the two following conditions are
satis
ed:
(i) The subspaces H1 and H2 are orthogonal to each other.
(ii) Each v 2 H can be written as v = v1 + v2 with v1 2 H1 and v2 2 H2 .
Ex. 1.22 Let H1 and H2 be Hilbert spaces and make H := f(v1
v2) j v1 2 H1
v2 2 H2 g
into an inner product space by the rules
(v1
v2) + (w1
w2) := (v1 + w1
v2 + w2 )
(v1
v2 ) := (v1
v2)

h(v1
v2)
(w1
w2)i := hv1
w1i + hv2
w2i:
Show that H is a Hilbert space and that it is the direct sum of its two closed linear
subspaces f(v
0) j v 2 H1 g and f(0
v) j v 2 H2 g.
An example of a direct space decomposition is given by the following Proposition.
1.23 Proposition The space L22 is the orthogonal direct sum of the two closed linear
subspaces
L22 even := ff 2 L22 j f (t) = f (;t) a:e:g

L22 odd := ff 2 L22 j f (t) = ;f (;t) a:e:g:
The maps f 7! f j0 ]: L22 even ! L2(0
]) and f 7! f j0 ]: L22 odd ! L2(0
]) are
isomorphisms of Hilbert spaces, provided the inner product on L2 (0
]) is normalized as
Z
1
hf
gi :=  f (t) g(t) dt:
(1:11)
0
p
Theorem The functions t 7! 1 and t 7! 2 cos(nt) (n = 1
2
: : :) form an orthonormal
basis pof L22 even, and hence also of L2(0
]) (with inner product (1.11)). The functions
t 7! 2 sin(nt) (n = 1
2
: : :) form an orthonormal basis of L22 odd, and hence also of
L2(0
]) (with inner product (1.11)).

L2 THEORY

Ex. 1.24 Give the proofs of the above Proposition and Theorem.

1.5 Further exercises

Ex. 1.25 Let f be the sawtooth function, i.e. the 2-periodic function which is given on
(0
2) by:

f (x) :=  ; x (0 < x < 2)

and which is arbitarary for x = 0. Show the following:
(a) f 2 L22 and fb(n) = (in);1 if n 6= 0 and fb(0) = 0.
P
;2
2
(b) 1
n=1 n =  =6.
Hint Apply formula (1.9).

Ex. 1.26 Show that

(1:12)

1
X
n;4 = 4=90

n=1

by applying formula (1.9) to the function f 2 L22 such that f (t) := t2 for t 2 (;
).

Ex. 1.27 For 0 6= a 2 R let fa 2 L22 such that fa (t) := eat for t 2 (;
).
(a) Compute the Fourier coecients of fa .
(b) Let a
b 2 R such that a
b
a + b 6= 0. Derive the identity


1
X
1
1
1
coth(a) + coth(b) = 
+
n=;1 a ; in b + in
by applying (1.10) to the functions f := fa and g := fb.
(c) Show that the case a = b of this identity yields
N
X
1 :
;
1
coth(a) =  Nlim
!1 n=;N a ; in

PN
Does
P1 the right hand side converge? Is it allowed to replace limN !1 n=;N by
n=;1 ?

p
p
Ex. 1.28 Put c0 (t) := 1, cn (t) := 2 cos(nt) (n = 1
2
: : :), sn(t) := 2 sin(nt) (n =

1
2
: : :), and let

Z
1
hcm
sni :=  cm (t) sn(t) dt (m = 0
1
2
: : :
n = 1
2
: : :)
0
(inner products in L2 (0
])). Consider the 1
1 matrix with (real) matrix entries
hcm
sni, (row indices m and column indices n). Show that the columns of this matrix are
orthonormal, and that also the rows are orthonormal, i.e.,
1
1
X
X
hcm
sk ihcm
sli = kl

hck
snihcl
sni = kl :
m=0

Next compute the matrix entries explicitly.

n=1

10

CHAPTER 1

Ex. 1.29 Let f 2 L22 . For k = 1
2
: : : put gk (t) := f (kt). Then gk 2 L22 . Express the
Fourier coecients of gk in terms of the Fourier coecients of f .

Ex. 1.30 Let
 2 f1g. De
ne

L22  := ff 2 L22 j f (;x) = f (x) a:e:
f ( ; x) =  f (x) a:e:g:
a) Show that the Hilbert space L22 is the direct sum of the four mutually orthogonal
closed linear subspaces L22  (
 2 f1g).
b) For each choice of

nd an orthonormal basis of L22  . (Start with the orthonormal basis for L22 even or L22 odd given in x1.23.)
c) For each
 give a Hilbert space isomorphism of L22  with L2 (0
12 ]).

Ex. 1.31 For F a function on R de
ne a function f on (0
2) by f (x) := F (cot 21 x).

This establishes a one-to-one linear correspondence between functions f on (0
2) and
functions F on R
(a) Show that the map f 7! F is a Hilbert space isomorphism of L2((0
2) (2);1 dx)
onto L2 (R  ;1 (t2 + 1);1 dt)., i.e.,
1 Z jf (x)j2 dx = 1 Z 1 jF (t)j2 dt :
2 ;
 ;1
t2 + 1
(b) Show that the orthonormal basis of L2 ((0
2) (2);1 dx) provided by the functions
x 7! einx (n 2 Z), is sent by the map f 7! F to the orthonormal basis of L2(R  ;1 (t2 +
1);1 dt) given by the functions t 7! ( tt;+ii )n .

Ex. 1.32 Let V be the linear space of piecewise linear continuous functions on the closed
bounded interval a
b]. So, f 2 V i f 2 C (a
b]) and there is a partition a = a0 < a1 <

a2 : : : < an = b such that the restriction of f to ai;1
ai] is linear for i = 1
: : :
n. Let
V0 := ff 2 V j f (a) = f (b) = 0g. Show that V is dense in C (a
b]) and that V0 is dense
in Lp (a
b]) (1
p < 1).

11

2 L1 theory
2.1 Growth rates of Fourier coe cients
2.1 The Fourier coecients fb(n) (n 2 Z) of a function f 2 L12 were de
ned in formula
(1.8). It follows from this formula that
jfb(n)j
kf k1 (f 2 L12
n 2 Z):
Hence
kfbk1
kf k1 where kfbk1 := sup jfb(n)j:

(2:1)

n2Z

So fb 2 `1 (Z) if f 2 L12 . Here `1 (Z) := f(n)n2Z j supn2Z jn j < 1g, a Banach space.
Ex. 2.2 Show that the map f 7! fb: L12 ! `1(Z) is a bounded linear operator. Also
determine its operator norm.
2.3 In x1.15 we gave the so-called Riemann-Lebesgue Lemma in the L2 -case. It remains
valid for the L1-case:
Theorem (Riemann-Lebesgue Lemma) If f 2 L12 then limjnj!1 fb(n) = 0.
Proof Let f 2 L12 . Let " > 0. We have to look for a natural number N such that
jfb(n)j < " if jnj  N . Since L22 is dense in L12 (see x1.10), there exists a function g 2 L22
such that kf ; gk1 < 21 ". By the Riemann-Lebesgue Lemma for L2 (see x1.15) there exists
N 2 N such that: jnj  N ) jgb(n)j < 21 ". Hence, if jnj  N then
jfb(n)j
jfb(n) ; gb(n)j + jgb(n)j
kf ; gk1 + jgb(n)j < 21 " + 12 " = "

where we used the inequality (2.1).
Ex. 2.4 De
ne the linear space c0 (Z) by
c0 (Z) := ffcn gn2Z 2 `1 (Z) j jnlim
c = 0g:
j!1 n

Show that c0 (Z) is a closed linear subspace of `1 (Z). So, in particular, c0 (Z) becomes a
Banach space. Note also that f 7! fb: L12 ! c0 (Z) is a bounded linear map.
2.5 In addition to the space C2 we will deal, for k = 1
2
: : : or 1, with the linear
space C2k consisting of 2-periodic C k -functions on R . By C20 we will just mean C2 .
Elementary integration by parts yields:
(f 0 )b(n) = in fb(n) (f 2 C21
n 2 Z):
(2:2)
This is an important result. Corresponding to the dierentiation operator f 7! f 0 acting
on 2-periodic functions we have on the Fourier transform side the operator multiplying
a function of n by in. Iteration of (2.2) yields:
(f (k) )b(n) = (in)k fb(n) (f 2 C2k
k = 0
1
2
: : :
n 2 Z):
(2:3)
For f 2 C2 the Riemann-Lebesgue Lemma gives that fb(n) = o(1) as jnj ! 1, a
modest rate of decline for the Fourier coecients. It turns out that a higher order of
dierentiability of a 2-periodic function f implies a faster decline of fb(n) as jnj ! 1:

12

CHAPTER 2

Theorem (a) Let k 2 f0
1
2
: : : g. If f 2 C2k then fb(n) = o(jnj;k ) as jnj ! 1.

(b) If f 2 C21 then fb(n) = O(jnj;k ) as jnj ! 1 for all k 2 f0
1
2
: : : g.

So we see that for 2-periodic C 1 -functions the Fourier coecients decrease faster
in absolutele value to zero than any inverse power of jnj. We then say that the Fourier
coecients are rapidly decreasing.
Proof of Theorem Part (b) follows immediately from part (a). For the proof of part
(a) we use equations (2.3) and the Riemann-Lebesgue Lemma:
jfb(n)j = jnj;k j(f (k) )b(n)j = jnj;k o(1) as n ! 1,
since f (k) 2 C2 if f 2 C2k .
Ex. 2.6 Show that (2.2) still holds if f 2 C2 with piecewise continuous derivative.
Ex. 2.7 Let f 2 C2 . Suppose that f can be extended to an analytic function on a region
in C which contains the strip fz 2 C j jIm zj
K g. Show that jfb(n)j = O(e;K jnj ) as
jnj ! 1.
Hint If z 2 C and jIm zj
K then f (z + 2) = f (z). Next show by contour integration
that
Z a+
;inz
bf (n) = 1
2 a; f (z) e dz (n 2 Z
jIm aj
K ):
2.8 In the previous sections we derived the behaviour of the Fourier coecients fb(n) from
the behaviour of the function f . Now we consider the inverse problem: Let coecients n
with certain behaviour be given. Find a 2-periodic function f such that fb(n) = n and
give the behaviour of f .
We sayPthat
the doubly in
nite sequence (n )n2Z is in `1(Z) if
1
k(n)k1 := n=;1 jnj < 1.
Theorem Let (n ) 2 `1(Z). Put
1
X
f (x) :=
n einx (x 2 R )

(2:4)
n=;1

well de
ned because the series converges absolutely. Then f 2 C2 and fb(n) = n (n 2 Z).
Also kf k1
k(n)k1 .
Proof The absolute convergence of the series on the right hand side Pof1(2.4) is uniform
for x 2 R because of the Weierstrass test, since jn einx j
jn j and n=;1 jn j < 1.
The sum of a uniform convergent series of continuous functions is continuous. Since the
terms of the series are 2-periodic in x, the same holds for the sum function. Thus f 2 C2 .
The inequality kf k1
k(n )k1 follows by taking absolute values on both sides of (2.4),
and by dominating the absolute value of the sum on the right hand side by the sum of the
absolute values of the terms. Finally, we!derive that


Z
Z
1
1
X
X
1
1
i
(
m
;
n
)
x
imx
;
inx
m e e
dx =
m 2
e
dx = n

fb(n) = 2
;

m=;1

m=;1

where the second equality is permitted because the integral over a bounded interval of a
uniformly convergent sum of continuous functions equals the sum of the integrals of the
terms.

L1 THEORY

13

Ex. 2.9 Let f (x) be given by (2.4). Prove the following statements. Use for (a) and (b)

the theorem about dierentiation of a series of functions on a bounded interval for which
the series of derivatives is uniformly convergent.
P
(a) If 1
then f 2 C21 .
n=;1 jnj jnj < 1 P
k
k
(b) Let k 2 f0
1
2
: : :g. If 1
n=;1 jnj jn j < 1 then f 2 C2 .
(c) Let  > 1. If n = O(jnj; ) as jnj ! 1 then f 2 C2k for all integer k such that
0
k <  ; 1.
(d) If n = O(jnj;k ) as jnj ! 1 for all k 2 f0
1
2
: : : g then f 2 C21 .
(e) Let K > 0. If n = O(e;K jnj ) as jnj ! 1 then f can be extended to an analytic
function on the strip fz 2 C j jIm zj
K g.

2.10 Remark Observe that the Fourier images of L22 and of C21 can be completely

characterized. Namely, for a given sequence (n)n2Z we have:

P
2
n = fb(n) (n 2 Z) for some f 2 L22 i 1
n=;1 jn j < 1
n = fb(n) (n 2 Z) for some f 2 C21 i n = O(jnj;k ) as jnj ! 1 for all k 2 f0
1
2
: : : g.
However, such a characterization is not possible for the Fourier images of C2
C21
C22
: : :
and of L12 .

Ex. 2.11 Let 0 := 0 and n := jnj; for n 2 Znf0g. Then it is known in the literature
for which real values of  there exists f 2 L12 such that fb(n) = n (n 2 Z). For some

values of  this follows from a non-trivial theorem (see sections 7.17 and 7.19 in van Rooij).
However, for some other values of  the answer is immediate. Give the answer in these
cases.

2.2 Fubini's Theorem

2.12 For the treatment of convolution we will need Fubini's Theorem. The Proposition

below formulates Fubini's theorem for nonnegative measurable functions, see also syll.
Integratietheorie, Sections 5.6{5.8. Next the Theorem below will deal with complex-valued
measurable functions. Below we will work with -
nite measure spaces (X
A
) and
(Y
B
 ) and their product space (X
Y
A B

 ), again a -
nite measure space.
It is helpful to read
rst the Theorem below for the case that X and Y are bounded
intervals,  and  are Lebesgue measure, and the function h: X
Y ! C is continuous.
Then all integrals can also be considered as Riemann integrals and the assumption in the
Theorem concerning the absolute convergence of a repeated integral is automatic. For this
case the Theorem was already treated in syll. Analyse B2.
The general Proposition and Theorem below are much more delicate, because measure
spaces may be -
nite rather than
nite and because measurable rather than continuous
functions are considered.

2.13 Proposition Let the function h: X

Y ! 0
1] be measurable with respect to
the -algebra A B. Then:
R
(a) The function y 7! RX h(x
y) d(x) is measurable on B
(b) The function x 7! Y h(x
y) d (y) is measurable on A

14

CHAPTER 2

(c) We have
Z
Z
h(x
y) d(
 )(x
y) =

h(x
y) d(x) d (y):

(2:5)
Note that all integrals considered in (2.5) are integrals of measurable functions which
take values on 0
1]. The integral of such a function is well-de
ned and it will yield a
number in 0
1].
Consider next a function h: X
Y ! C (so not necessarily non-negative) which is
measurable with respect to the -algebra A B. Then the nonnegative function jhj is
measurable with respect to A B, so the above Proposition will hold with h(x
y) replaced
by jh(x
y)j. This we will need in the next theorem.
2.14 Theorem (see Rudin, Theorem 7.8) Let the function h: X
Y ! C be measurable with respect to the -algebra A B. Suppose that one of the two inequalities below
is valid.


Z Z
jh(x
y)j d(x) d (y) < 1:
(2:6)
Y
X


Z Z
jh(x
y)j d (y) d(x) < 1:
(2:7)
X Y

h(x
y) d (y) d(x) =

(So the
R other inequality is also valid by the above Proposition.) Then:
(a) X jh(Rx
y)j d(x) < 1 for y outside some set Y0  Y of  -measure 0 and the function
y 7! X h(x
y) d(x): Y nY0 ! C , arbitrarily extended to a complex-valued function
Ron Y , is  -integrable on Y .
(b) Y jh(Rx
y)j d (y) < 1 for x outside some set X0  X of -measure 0 and the function
x 7! Y h(x
y) d (y): X nX0 ! C , arbitrarily extended to a complex-valued function
on X , is -integrable on X .
(c) Formula (2.5) is valid.
The crucial condition to be checked in this Theorem is inequality (2.6) or (2.7). The
important conclusion in the Theorem is that (almost) all integrals in (2.5) are well-de
ned
and that the equalities in (2.5) (in particular the second one) hold.

2.3 Convolution
2.15 Denition The convolution product of two 2-periodic functions f
g is a function
f  g on R given by

Z
1
(f  g)(x) := 2
f (t) g(x ; t) dt (x 2 R )

(2:8)
;
provided the right hand side is well-de
ned. Then, clearly, f  g is also 2-periodic and
f  g = g  f . Check these two properties by simple transformations of the integration
variable in (2.8)
For deriving further properties of the convolution product, the easiest case is when
f
g are continuous:

L1 THEORY

15

2.16 Proposition If f
g 2 C2 then f  g 2 C2 and

(f  g)b(n) = fb(n) bg(n) (n 2 Z):
(2:9)
Proof For the proof of the continuity (in fact uniform continuity) of f  g observe that

 Z
1


jg(x ; t) ; g(y ; t)j:
2 ; f (t) (g(x ; t) ; g(y ; t)) dt
kf k1 sup
t2R

Let " > 0. Since any periodic continuous function is uniformly continuous on R (why?),
there exists  > 0 such that jg(x ; t) ; g(y ; t)j < " for all t 2 R if jx ; yj < . Hence
j(f  g)(x) ; (f  g)(y)j < " kf k1 if jx ; yj < .
For the proof of (2.9) use Fubini's theorem in the case of continuous functions on a
bounded interval. Thus


Z Z
1
(f  g)b(n) = 42
f (t) g(x ; t) dt e;inx dx
;
;


Z
Z
1
;
int
;
in
(
x
;
t
)
f (t) e
g(x ; t) e
dx dt
= 42
;
;
Z
1
= 42
f (t) e;int gb(n) dt = fb(n) bg(n):
;

Ex. 2.17 Let f (x) := eimx , g(x) := einx (m
n 2 Z). Compute f  g. Also check that the

result agrees with equality (2.9).

Ex. 2.18 Prove by Fubini's theorem in the case of continuous functions on a bounded
interval the following. If f
g
h 2 C2 then
(f  g)  h = f  (g  h) (associativity):
(2:10)
2.19 Denition-Theorem Let f
g 2 L12 . Let (f  g)(x) be de
ned by (2.8) for those
x 2 R for which the integral on the right hand side of (2.8) converges absolutely.
(a) For almost all x 2 R the integral on the right hand side of (2.8) converges absolutely.
Extend f  g to a (2-periodic) function on R by choosing arbitrary values of (f  g)(x)
on the set of measure zero where f (x) is not yet de
ned by (2.8). Then f  g 2 L12
and

kf  gk1
kf k1 kgk1:

(2:11)
(b) The equivalence class of f  g only depends on the equivalence classes of f and g. In
other words, for f
g 2 L12 the convolution product f  g is well-de
ned as an element
of L12 .
(c) For f
g 2 L12 , formula (2.9) is valid.
(d) For f
g
h 2 L12 , the associativity property (2.10) is valid.
Proof Let f
g 2 L12 . First observe that




1 Z Z jf (t) g(x ; t)j dx dt = 1 Z jf (t)j Z jg(x ; t)j dx dt
42 ; ;
42 ;
;
Z
1
= 2
jf (t)j kgk1 dt = kf k1 kgk1 < 1:
(2:12)
;

16

CHAPTER 2

Hence, by Fubini's Theorem 2.14 we have that

Z
jf (t) g(x ; t)j dt < 1
;

for almost all x. Thus, (f  g)(x) is well-de
ned by (2.8) for almost all x, and f  g is
measurable on R . Again by Theorem 2.14, it follows that


1 Z Z f (t) g(x ; t) dt dx < 1:

42 ;  ;
So f  g 2 L12 . For the proof of (2.11) use that


Z Z
Z
1
1
jf (t) g(x ; t)j dt dx = 42
kf  gk1
42
;
;
;

Z
;

jf (t) g(x ; t)j dx dt

by Proposition 2.13, and combine with (2.12).

For the proof of part (b) let f
g 2 L12 with kf k1 = 0 or kgk1 = 0. Then kf  gk1 = 0
because of (2.11).
For the proof of (c) repeat the proof of (2.9) for the continuous case. We can now
justify the interchange of integration order in the second equality of that proof by Fubini's
Theorem 2.14, because


1 Z Z jf (t) g(x ; t) e;inx j dx dt < 1:
42 ; ;
This last inequality is true because the left hand side equals k jf j  jgj k1 < 1.
Finally (d) can be proved similarly as in Exercise 2.18, with justi
cation of the interchange of integration order by Theorem 2.14.

2.20 Proposition If f 2 L12 and g 2 C2 then f  g 2 C2 and

kf  gk1
kf k1 kgk1:

(2:13)

Proof The proof given in Proposition 2.16 that f  g 2 C2 if f
g 2 C2 , still works if the
assumption on f is relaxed to f 2 L12 . The proof of inequality (2.13) is straightforward.
Ex. 2.21 Show: If f
g 2 L22 then f  g 2 C2 and

kf  gk1
kf k2 kgk2:

(2:14)

Hint First show that supx2R j(f  g)(x)j
kf k2 kgk2 if f
g 2 L22 . Next use this inequality
and the fact that f  g is continuous if f
g 2 C2 (see Proposition 2.16), together with the
density of C2 in L22 and the completeness of the Banach space C2 with respect to the
sup norm.

L1 THEORY

17

2.4 Further exercises

Ex. 2.22 Show that

1
X
2 cos x
2;n cos(nx) = 45 ;
; 4 cos x

n=0

(x 2 R ):

Ex. 2.23 Let A 2 C such that jAj 6= 1. Let f (x) := (A + eix );1 (x 2 R ). Compute the
Fourier coecients fb(n).

Ex. 2.24 Let f 2 L12 . Express gb(n) in terms of the fb(m) if:
(a) g(x) = f (x + a) (a 2 R )
(b) g(x) = f (;x)
(c) g(x) = f (x)
(d) g(x) = f (;x).

Ex. 2.25 Let f 2 L12 , g 2 C21 . Show that f  g 2 C21 and that (f  g)0 = f  g0.

18

3 The Dirichlet kernel

3.1 Denition of the Dirichlet kernel

3.1 Let f 2 L12 . The formal doubly X

in
nite series
n2Z

fb(n)einx

(3:1)

is called the Fourier series of f . We will see that this series converges in a certain sense to
f (x), but the type of convergence will depend on the nature of f . Saying that the series
(3.1) converges in some sense to f (x) amounts to the same as saying that the sequence of
partial Fourier sums
N
X
SN (x) = (SN f ])(x) :=
fb(n) einx (N = 0
1
2
: : :
x 2 R )
(3:2)
n=;N

converges in some sense to f (x) as N ! 1. Therefore it is important to examine the

functions SN f ] in more detail.
3.2 Denition-Proposition ZThe partial Fourier sum (3.2)Z can be written as
(SN f ])(x) = (f  DN )(x) = 21
f (t) DN (x ; t) dt = 21
f (x + t) DN (t) dt
(3:3)
;

where DN is the Dirichlet kernel:

8
N
< sin((N + 21 )x) (x 2= 2Z),
X inx >
DN (x) :=
e = > sin( 12 x)
: 2N + 1
n=;N
(x 2 2Z).

The Dirichlet Kernels D2 , D5 , D10 , centered at 0
2
4 respectively.

(See also the graph of DN in van Rooij, p.13.)

(3:4)

DIRICHLET KERNEL

19

Proof Observe from (3.2) that

 N
!
Z
X in(x;t)
1
f (t)
e
dt:
(SN f ])(x) = 2
;
n=;N
From (3.4) we see that DN 2 C2 and that it is an even function. The third equality in
(3.3) uses these last facts.

3.2 Criterium for convergence of Fourier series in one point

3.3 We will need the following three easy consequences of (3.4):

and

1 Z D (t) dt = 1

2 ; N
jDN (x)j
j sin(11 x)j
jxj (0 < jxj
)

2

(3:5)
(3:6)

1 ix
; 12 ix ;iNx
2
e
e
iNx
DN (x) =
e ;
e
(x 2= 2Z):
(3:7)
2i sin( 21 x)
2i sin( 12 x)
3.4 Theorem Let f 2 L12 , a 2 R . Suppose that one of the two following conditions
holds:
(a) There are M

 > 0 such that
jf (a + t) ; f (a)j
M jtj for ; < t < :
(3:8)
(b) f is continuous in a and also right and left dierentiable in a.
Then we have:
lim (S f ])(a) = f (a):
N !1 N
Remark Condition (a) of Theorem 3.4 implies continuity of f in a. Functions f satisfying
condition (a) are said to be Holder continuous of order in a. (The terminology Lipschitz
continuous of order is equally common.) If (b) holds then jf (a+tj)t;j f (a)j is bounded for t
in some neighbourhood of 0 since it has a limit as t # 0 and as t " 0. Thus if (b) holds then
(a) is satis
ed with = 1. So, we only need to prove the Theorem under condition (a).
Proof of Theorem 3.4 Assume condition (a). Let " > 0. We want to show that
jSN (a) ; f (a)j < " for N sucientlyZlarge. It follows from (3.3), (3.5) and (3.7) that
SN (a) ; f (a) = 21 (f (a + t) ; f (a)) DN (t) dt
Z ;
Z
iNt
=
ga+ (t)e dt ; ga; (t)e;iNt dt

(3:9)
;
;
where
 12 it
e
1
ga (t) = 2
(f (a + t) ; f (a)) (0 < jtj < ):
(3:10)
2i sin( 12 t)
For 0 < jtj <  we can estimate jga (t)j < 41 M jtj;1 (use (3.6) and (3.8)). For  < jtj < 
we can estimate jga (t)j < (4 sin( 21 ));1jf (a + t) ; f (a)j. Hence the functions ga are in
L1(;
]). By the Riemann-Lebesgue Lemma (Theorem 2.3) it follows that (3.9) tends
to 0 as N ! 1.

20

CHAPTER 3

3.5 Theorem Let f 2 L12 , let a 2 R , and suppose that the limits
f (a+ ) := lim
f (x)

x#a

f (a;) := lim
f (x)
x"a

exist. Suppose that one of the two following conditions holds:

(a) There are M

 > 0 such that

jf (a + t) ; f (a+)j
Mt and jf (a ; t) ; f (a;)j
Mt for 0 < t < :

(b) f is right and left dierentiable in a in the sense that the following two limits exist:

f (a + t) ; f (a+)

f 0(a+ ) := lim
t#0
t
Then we have:

f (a ; t) ; f (a;) :
f 0 (a;) := lim
t#0
;t

1 (f (a+) + f (a; )):

lim
(
S

f
])(
a
)
=
N
2
N !1

Proof Since DN is an even function, we conclude from (3.3) that

Z
1
1
SN (a) = 2
2 (f (a + t) + f (a ; t)) DN (t) dt:
;

Hence

SN (a) ;

1 (f (a+ ) + f (a; )) =
2

1 Z
2 ;

f (a + t) + f (a ; t) ; f (a+) + f (a;) D (t) dt:
N
2
2

Now put g(t) := 21 (f (a + t) + f (a ; t)) for 0 < jtj
 and g(0) := limt!0 g(t) =
1 (f (a )+ f (a )). Then condition (a) or (b) for f at the point a implies the corresponding
+
;
2
condition in Theorem 3.4 for g at the point 0. Now apply Theorem 3.4 to g at 0.

3.6 Corollary (localization principle) Let f
g 2 L12 , let a 2 R , and suppose that

f (x) = g(x) for x in a certain neighbourhood

; of a. Then

1 precisely
; one of the

1following two
alternatives holds for the two sequences (SN f ])(a) N =0 and (SN g])(a) N =0 :
(a) The two sequences both converge and have the same limit.
(b) The two sequences both diverge.
Proof Apply Theorem 3.4 to the function f ; g at a. Since f ; g is identically zero in a
neighbourhood of a, we conclude that limN !1 (SN f ; g])(a) = 0. Hence
;

lim
(
S

f
])(
a
)
;
(
S

g
])(
a
)
= 0:
N
N
N !1
Thus the convergence and possible sum of the Fourier series of a function f 2 L12 at
a point a is completely determined by the restriction of f to an arbitrarily small neighbourhood of f . If we leave a given function f unchanged on such a neighbourhood, but
change it elsewhere, then the Fourier coecients fb(n) may become completely dierent,
but convergence or divergence and the possible sum of the Fourier series at a will remain
the same. This explains why the above Corollary is called localization principle.

DIRICHLET KERNEL

21

Ex. 3.7 Let f be the sawtooth function de
ned in Exercise 1.25, formula (1.12). Put
f (x) := 0 for x 2 2Z.
(a) Prove that

1 sin(nx)
X
f (x) = Nlim
(S f ])(x) = 2
!1 N
n=1 n

(3:11)

with pointwise convergence for all x 2 R .

(b) Show by substitution of x := =2 or x := =4 in (1.12), (3.11) that

 = 1 ; 1 + 1 ; 1 + 

4
3 5 7

p = 1 + 31 ; 15 ; 17 + 91 + 111 ; 131 ; : : : :
8

(3:12)
(3:13)

Ex. 3.8 Let 0 <  < . Let f be a 2-periodic function which is given on ;
] by:



],
f (x) := 10 ifif x <2 j;
xj
 .
(a) Determine fb(n) (n 2 Z).
(b) Determine

N sin(n )
X
einx
lim
N !1 n=;N n

provided the limit exists.

Ex. 3.9 Let f be a 2-periodic function which is C 1 outside 2Z and such that f (0+ ),
f (0;), f 0(0+ ), f 0(0; ) exist in the sense of Theorem 3.5. Prove that fb(n) = O(jnj;1) as
jnj ! 1, but not necessarily fb(n) = o(jnj;1) as jnj ! 1.
Hint Use Theorem 2.5 and Exercise 1.25(a).

Ex. 3.10 Let f be a 2-periodic function which is C 1 outside a discrete set X having the
property that X \ ;
) is a
nite set. Assume that f (x+ )
f (x;)
f 0(x+)
f 0(x; ) exist
in the sense of Theorem 3.5 for each x 2 X . Prove that fb(n) = O(jnj;1) as jnj ! 1.

3.3 Continuous functions with non-convergent Fourier series

We will now show in a \soft" way (i.e., by using functional analytic methods) that there
exists a 2-periodic continuous function for which the Fourier series does not converge
everywhere. For this we need a \hard" estimate of the Dirichlet kernel in the two lemmas
below. This estimate will also be useful elsewhere. We refer to van Rooij, x3.18 for
an explicit example of a 2-periodic continuous function with not everywhere converging
Fourier series.

22

CHAPTER 3

3.11 Lemma Let

Then

(x) := cot( 21 x) ; 2x;1 (0 < jxj < 2):

(3:14)

DN (x) = 2 sin(xNx) + (x) sin(Nx) + cos(Nx) (0 < jxj < 2):

(3:15)
Proof It follows from (3.4) that, for 0 < jxj < 2,
sin((N + 21 )x)
= sin(Nx) cot( 12 x) + cos(Nx):
DN (x) =
sin( 21 x)
Ex. 3.12 Show the following:
(a) limx!0 (x) = 0
(b) limx!0 0(x) = ; 16 
(c) Put (x) := 0. Then  is dierentiable and strictly decreasing on (;2
2).
(d) () = ;2;1 , (;) = 2;1, sup; x j(x)j = 2;1 .

3.13 Lemma We have

Z
jDN (x)j dx = 8;1 log N + O(1) as N ! 1:
;

In particular,

lim

N !1 ;

jDN (x)j dx = 1:

Proof It follows from (3.15) and Exercise 3.12(d) that



jDN (x)j ; j2x;1 sin(Nx)j
jDN (x) ; 2x;1 sin(Nx)j
2;1 + 1 (0 < jxj < ):
Hence

Z  sin(Nx) 
jDN (x)j dx = 2  x  dx + O(1) as N ! 1:
;
;

Next we write
Z  sin(Nx) 
Z N j sin(x)j


2  x  dx = 4
x dx
;
0
NX
;1 Z 1 sin(x)
dx
=4
n=0 0 x + n
NX
;1 Z 1 sin(x)
=4
dx + O(1) as N ! 1
x
+
n
0
n=1
;1
N Z 1 cos(x)
X
(1) ;1 NX
;
1
;
1
;
1
= 4
(n + (n ; 1) ) ; 4
2 dx + O(1) as N ! 1
n=1
n=1 0 (x + n)
N
(2) ;1 X
= 8
n;1 + O(1) as N ! 1
(3)

n=1

= 8;1 log N + O(1) as N ! 1.

DIRICHLET KERNEL

23

R1
In equality (1) we used integration by parts.
In
equality
(2)
we
majorized
j
0 (x +
PN ;2
;
2
;
2
n) cos(x) dxj by
we used that n=1 n = O(1) as N ! 1. In equality
PNn and
;
1
(3) we used that n=1 n = log N + O(1) as N ! 1 (by comparing with the corresponding Riemann integral, see the de
nition of Euler's constant  in Syll. Analyse A3).

Ex. 3.14 Let : ;
] ! R be continuous with only
nitely many zeros. De
ne the
linear functional L: C2 ! C by
Z
1
f (x) (x) dx (f 2 C2 ):
L(f ) := 2
;

Prove that L is a bounded linear functional with operator norm

Z
1
kLk = kk1 = 2 j(x)j dx:
;

(3:16)

(3:17)

3.15 Theorem For all x 2 R there exists f 2 C2 such that the sequence (SN f ])(x) 1N =0
does not converge to a
nite limit.
Proof Without loss of generality we may take x := 0. De
ne the linear functional
LN : C2 ! C by
Z
1
LN (f ) := (SN f ])(0) = 2
f (t) DN (t) dt:
;

It follows from Exercise 3.14 that kLN k = kDN k1 , and it follows next from Exercise 3.13
that the sequence (kLN k)1
N =0 is unbounded. Now suppose1that limN !1 LN (f ) exists for
all f 2 C2 . Then, for all f 2 C2 , the sequence (LN (f ))N =0 is bounded. Hence, by the
Banach-Steinhaus theorem (see syll. Functionaalanalyse, x5.5), the sequence (kLN k)1
N =0
is bounded. This is a contradiction.
;

Ex. 3.16 Let f 2 L12 and let a 2 R . Prove that for the two sequences (SN f ])(a) 1N =0
and
1 Z f (a + t) sin(Nt) dt
N = 0
1
2
: : :

t
;

precisely one of the following two alternatives holds:

(a) Both sequences converge with the same limit.
(b) Both sequences diverge.
Conclude that, for f satisfying one of the conditions of Theorem 3.5, we have
Z
sin(Nt) dt:
1 (f (a+) + f (a; )) =  ;1 lim
f
(
a
+
t
)
2
N !1 ;
t

(3:18)

Ex. 3.17 Prove, by using (3.18), that

Z sin(Nt)
Z 1 sin x
dx = Nlim
dt = 12 :
!1
x
t
0
0

(3:19)

24

CHAPTER 3

3.4 Injectivity of the Fourier transform

3.18 In this subsection we will prove that a function f 2 L12 which has all its Fourier
coecients fb(n) = 0, is almost everywhere equal to zero. For the case that f 2 L22 , this

was already implied by Parseval's identity (1.9). However, Parseval's identity depended
on Theorem 1.11, for which we postponed the proof. Let us
rst consider the case that
f 2 C2 .

Proposition Let f 2 C2 such that fb(n) = 0 for all n 2 Z. Then f = 0.

R
Proof Let F (x) := 0x f (t) dt. Then F is a C 1 -function on R and, because fb(0) = 0,
the function F is 2-periodic. Thus F 2 C21 . Since F 0 = f , we have fb(n) = in Fb(n) (see
(2.2)). Hence Fb(n) = 0 if n =
6 0. It follows that (SN F ])(x) = Fb(0). By Theorem 3.4 we

conclude that F (x) = limN !1 (SN F ])(x) = Fb(0). Hence f (x) = F 0 (x) = 0.
For the similar result in the case that f 2 L12 we proceed as follows.
R
3.19 Lemma Let f 2 L12 such that fb(0) = 0. Put F (x) := 0x f (t) dt. Then F 2 C2
and fb(n) = in Fb(n).
Proof Continuity of F is a consequence of the dominated convergence theorem, applied
to the functions fxt, where xt is the characteristic function of the interval x
x + t)
and letting t ! 0. Now we compute for n 6= 0:
Z2 Zx
Z2 Z2
;
inx
b
F (n) =
f (t)e dt dx =
e;inx dxf (t) dt
0
t
Z02 0e;int ; 1

1 (fb(n) ; fb(0)) = 1 fb(n):
=
f
(
t
)
dt
=
in
in
in
0
We used Fubini's theorem for the second equality.
Remark It is a non trivial result of Lebesgue that for almost all x F 0 (x) exists and
equals f (x). (see Rudin, Theorem 8.17)

3.20 Theorem (a) Let f 2 L12 such that fb(n) = 0 for all n 2 Z. Then f = 0 a.e. .
(b) Let f
g 2 L12 such that fb(n) = gb(n) for all n 2 Z. Then f = g a.e. .
Proof It is sucient to prove part (a). Let F be as in Lemma 3.19. Then it follows
that F 2 C2 and that Fb(n) = 0 for n =
6 0. Thus F ;R Fb(0) = 0 by Proposition 3.18.
Since F (0) = 0, we Rconclude that F = 0. It follows that ab f (t) dt = 0 for every choice of
a
b 2 R . But then E f (t) dt = 0 for (bounded) open and
R also for (bounded) closed sets.
By regularity of Lebesgue measure, we conclude that E f dt = 0 for every Borel set. It
follows that f = 0 a.e. .

3.21 Corollary (a) The linear span of the functions t 7! eint (n 2 Z) is dense in L22
with respect to the norm k : k2.
(b) The functions t 7! eint (n 2 Z) form an orthonormal basis of L22 .
Proof Apply De
nition-Theorem 1.12 to the case f 2 L22 of Theorem 3.20.

DIRICHLET KERNEL

25

So we have now proved parts (b) and (c) of Theorem 1.11, therefore we have also
de
nitely established all results in subchapter 1.2 which depended on Theorem 1.11, see
the summarizing x1.18. In the next chapter we will also prove part (a) of Theorem 1.11.
As a corollary of the injectivity result of Theorem 3.20 we can also give the following
addition to Theorem 2.8.

3.22 Corollary Let f 2 L12 . If fb 2 `1(Z) then f 2 C2 and, for almost all x 2 R :
1
X
f (x) =
fb(n) einx :
n=;1

(3:20)

The right hand side of (3.20) converges absolutely and uniformly for x 2 R .
Proof Denote the right hand side of (3.20) by g(x). Then, by 2.8, g 2 C2 and gb(n) =
fb(n) for all n 2 Z. Now apply Theorem 3.20(b).

3.5 Uniform convergence of Fourier series

We will now consider an analogue of Theorem 3.4 such that f satis
es not just a Holder
condition at one point, but a uniform Holder condition on some interval (a
b). Then it
will turn out that SN f ] converges uniformly to f on each compact subset of (a
b). For
f 2 C22 this result is almost immediate:

3.23 Proposition Let f 2 C22 . Then limN !1 SN f ] = f , uniformly on R .

Proof It follows from Theorem 2.5(a) that fb(n) = o(jnj;2) as jnj ! 1. Now apply
Corollary 3.22.

3.24 The following lemma quickly follows from the well-known Arzela-Ascoli theorem
(see for instance A. Browder, Mathematical Analysis, Springer, 1996, Theorem 6.71 and
Corollary 6.73).

Lemma Let (X
d) be a compact metric set. Let (n )1n=1 be a sequence of complex-

valued functions on X which is equicontinuous on X , i.e., such that, for each " > 0, there
is a  > 0 with the property that jn (x) ; n (y)j < " for al n 2 N if x
y 2 X and
d(x
y) < . Suppose that limn!1 n (x) = 0 for x 2 X . Then limn!1 n = 0 uniformly
on X .
Proof Suppose that the convergence n ! 0 is not uniform on X . Then there exist " > 0,
an increasing sequence of positive integers n1
n2
: : :, and a sequence x1
x2
: : : in X such
that jnk (xk )j  ". By compactness of X the sequence x1
x2
: : : has a subsequence
converging to some x0 2 X . Without loss of generality we may assume that the sequence
x1
x2
: : : already converges to x0. Then

jnk (xk )j
jnk (xk ) ; nk (x0)j + jnk (x0 )j:

By the assumptions, there exists K 2 N such that jnk (x0 )j < 21 " if k  K and jm (xk ) ;
m (x0)j < 12 " for all m 2 N if k  K . Hence nk (xk ) < " if k  K . This is a contradiction.

26

CHAPTER 3

3.25 Theorem Let f 2 L12 . Suppose that there exist an interval (a
b) and positive
real numbers M
such that

jf (x) ; f (y)j
M jx ; yj for all x
y 2 (a
b)

(3:21)

(a uniform Holder condition on (a
b)). Then limN !1 SN f ] = f uniformly on each subinterval c
d] with a < c < d < b.
Proof By the proof of Theorem 3.4 we can write
Z
Z
iNt
(3:22)
SN (x) ; f (x) = gx+(t)e dt ; gx; (t)e;iNt dt
;

with the functions gx being given by (3.10). Put

Z
N (x) := gx(t) eiNt dt (x 2 c
d]):
;

By (3.21) and the proof of Theorem 3.4, the functions gx belong to L12 for x 2 c
d]. So
limN !1 N (x) = 0 for x 2 c
d] by the Riemann-Lebesgue Lemma (Theorem 2.3). We
will show that the functions N are equicontinuous on c
d]. Then we can apply Lemma
3.24 and thus prove the Theorem.
Let 0 <  <  such that  < c ; a and  < b ; d. Then we can estimate for x 2 c
d]
and 0 < jtj <  that jgx (t)j < 14 M jtj;1 (use (3.21) and (3.6)). We can estimate for
x
y 2 c
d] and  < jtj <  that
jgx(t) ; gy(t)j < (4 sin( 21 ));1 ;jf (x + t) ; f (y + t)j + jf (x) ; f (y)j
:
Next, for x
y 2 c
d]:

jN (x) ; N (y)j

Z
jtj<

<jtj<

jgx(t) ; gy (t)j dt

Z
Z
) ; f (y)j :
1

;
1

M jtj dt + 4 sin( 1 ) jf (x + t) ; f (y + t)j dt + jf (2xsin(
1 )
;
0
2
2
Let " > 0. We will
nd  > 0 such that theR last expression is dominated by " if x
y 2 c
d]
and jx ; yj <  . First take  such that M 0 jtj;1 dt < "=3. Then we can
nd  > 0 such
that the second and third term are dominated by "=3 if jx ; yj <  . For the third term
this follows because f is uniformly continuous on c
d]. For the second term this follows
from Proposition 1.9. This settles the equicontinuity of the functions N on c
d]. Thus
we can apply Lemma 3.24 and the Theorem will follow.

3.26 Corollary Let f 2 C21 , or let f 2 C2 with piecewise continuous derivative. Then
limN !1 SN f ] = f , uniformly on R .

Ex. 3.27 Let the 2-periodic function f be determined by f (x) := jxj for ;
x
.
Determine the Fourier coecients fb(n). The uniform convergence of (SN f ])1
N =0 on R is
implied by Corollary 3.26. Check this uniform convergence independently by using the
explicit values of fb(n).

DIRICHLET KERNEL

27

3.6 The Gibbs phenomenon

The approximation of a piecewise continuous 2-periodic function (for instance the sawtooth function) by its partial Fourier sum shows a remarkable \overshooting" behaviour
near the jumps of the function. This behaviour is called the Gibbs phenomenon.

3.28 Let the 2-periodic function f be given by f (x) :=  ; x for 0 < x < 2. This

function was studied in Exercises 1.25 and 3.7. We have

N sin(nx) Z x
X
= DN (t) dt ; x
SN (x) = (SN f ])(x) = 2
n
0
Z xn=1
Z x sin(Nt)
;1 sin(Nx) ; x (0 < jxj < 2 )
(3:23)
dt
+

(
t
)
sin(
Nt
)
dt
+
N
=2
t
0
0
where we used (3.15) and where the function  is given by (3.14).

Ex. 3.29 R Let

(a
b) be a bounded interval and let c 2 (a
b). Let g 2 C 1 ((a
b)) and de
ne
x
Gn (x) := c g(t) e;int dt (n 2 Z, x 2 (a
b)). Prove that Gn (x) = O(jnj;1) as jnj ! 1,
uniformly for x in a compact subinterval of (a
b).
Hint Use integration by parts.

3.30 Let 0 < r < 2. We conclude from Exercises 3.29 and 3.12 that
Zx
0

(t) sin(Nt) dt = O(N ;1) as N ! 1, uniformly for jxj
r.

(3:24)

The integral sine is the function de
ned by

Z x sin y
Si(x) :=
dy (x 2 R ):
(3:25)
0 y
This is an even C 1 -function (in fact it is an analytic function). It follows from (3.19) that
1
Si(1) := xlim
!1 Si(x) = 2 :

The absolute maximum of Si(x) for x 2 0
1) is reached for x = . A numerical computation shows that, approximately,
Si()  1:179 : : : :
Si(1)
Let 0 < r < 2. It follows from (3.23), (3.25) and (3.24) that
)
SN (x) ; f (x) = 2Si(Nx) ;  + O(N ;1 )
as N ! 1, uniformly for 0 < x
r.
SN (;x) ; f (;x) = ;2Si(Nx) +  + O(N ;1 )
This implies:

;
;1  ) ; f (N ;1  )
= 2Si( ) ;   
0:179 : : :

lim
S
(
N
N
N !1
;
;1  ) ; f (;N ;1  )
= ;2Si( ) +   ;
0:179 : : : :
lim
S
(
;
N
N
N !1

28

CHAPTER 3

Also, for 0 <  < , we have

lim sup (S (x) ; f (x)) = 2Si() ; 

N !1 0<x N

(3:26)

lim
inf (S (x) ; f (x)) = ;2Si() + 

N !1 ; x<0 N

(3:27)

but

by Theorem 3.25.

lim
sup jS (x) ; f (x)j = 0
N !1 x 2 ; N

Ex. 3.31 Let h 2 C 1 (0
2]) Let g be a 2-periodic function which coincides with h
on (0
2). Write g(0+) := h(0) and g(0;) := h(2). Assume that g(0+)  g(0;). Let
0 <  < . Prove that

lim sup ((S g])(x) ; g(x)) = 21 (g(0+) ; g(0;)) (2;1Si() ; 1)

N !1 0<x N


lim
inf ((S g])(x) ; g(x)) = ; 12 (g(0+) ; g(0;)) (2;1Si() ; 1):
N !1 ; x<0 N

Hint Let the 2-periodic function f be given by f (x) :=  ; x for 0 < x < 2. Put
p(x) := g(x) ; g(0+) 2; g(0;) f (x) (0 < x < 2):
Then p 2 C2 with piecewise continuous derivative. Now use (3.26), (3.27), and Corollary
3.26.

The Gibbs phenomenon.

Left: the sawtooth and S64, Right: (SN (N ;1x) ; f (N ;1x))= for N = 1024

DIRICHLET KERNEL

29

3.7 Further exercises

Ex. 3.32 Prove that
sin x + 13 sin 3x + 15 sin 5x +    = =4 if 0 < x < .
Hint Use Exercise 3.8 or use (1.12), (3.11).

Ex. 3.33 Let  2 L12 . De
ne the linear functional L: C2 ! C by (3.16). Prove that L
is a bounded linear functional with norm given by (3.17).
Hint Use Exercises 3.14 and 1.32.

Ex. 3.34 Prove that DN (x) can be written as a polynomial of degree 2N in cos( 21 x).
Ex. 3.35 Prove that between each two successive zeros of DN (x) there is precisely one
zero of DN0 (x).

Ex. 3.36 1) Let fan g11 , fang11 be sequences of complex numbers. Prove (for N =
1
2
: : :) the Summation by parts formula
N
X
n=1

an (bn+1 ; bn ) = aN +1 bN +1 ; a1b1 ;

N
X
n=1

bn+1 (an+1 ; an ):

P
2) Write down a version of this formula for the case where bn = nj=1 cj .
3) Give a direct proof of the convergence of the series in exercise 3.32 (without using
that this is a Fourier series of a smooth function)
4) Show that the series
1 sin nx
X
n=2 log n
converges pointwise for every x.
5) What can you say about uniform convergence?

30

4 The Fejer kernel

4.1 Cesaro convergence

4.1 Denition Let (an )1n=1 be a sequence of complex numbers. If the sequence does

not have a limit then we may try to
nd a limit in a generalized sense by considering a
new sequence (bn )1
n=1 with bn being the mean of the
rst n elements of the sequence (ak ),
i.e.,
bn := n;1 (a1 + a2 +    + an ):
If a := limn!1 bn exists as a
nite limit then we say that the sequence (an) is Cesaro
convergent with Cesaro limit a. In that case we use the notation
(C) nlim
(4:1)
!1 an = a:
P
Similarly, we say that the series 1
nP=1 an is Cesaro convergent with Cesaro sum s if
the sequence (sn ) of partial sums sn := nk=1 ak is Cesaro convergent with Cesaro limit s.
In that case we use the notation
1
X
(C) an = s:
(4:2)
n=1
P
More generally, Cesaro convergence of the sum 1
Cesaro sum s means Cesaro
n=n0 an withP
1
convergence of the sequence (sn )n=n0 of partial sums sn := nk=n0 ak to Cesaro limit s,
i.e., that limn!1 (n ; n0 + 1);1 (sn0 +    + sn ) = s.

4.2 Example

The sequence 1
0
1
0
1
0
: : : does not converge but has Cesaro limit 12 .

The sum 1 ; 1 + 1 ; 1 + 1 ; 1 +    does not converge but has Cesaro sum 21 .

Ex. 4.3 Show the following. The sequence a0
a1
a2
: : : has Cesaro limit a i the

sequence a1
a2
: : : has Cesaro limit a.

Conclude that it is not necessary in the notation (4.1) to mention where the sequence (an)
starts.

4.4 Proposition If the sequence (an )1n=1 has limit a then it has also Cesaro limit a.
Proof Let M
N 2 N with M < N . For n  N we have
jn;1 (a1 +    + an ) ; aj
n;1 (ja1 ; aj +    + jaM ; aj) + n;1 (jaM +1 ; aj +    + jan ; aj)

MN ;1 sup jak ; aj + sup jak ; aj:
k=12:::

k
M +1

Let " > 0. Because the sequence (an ) converges, we have A := supk=12::: jak ; aj < 1
and we can choose M such that jak ; aj < 21 " if k  M . Hence

jn;1 (a1 +    + an ) ; aj
N ;1 MA + 12 " if n  N > M:

Choose N > M such that N ;1MA < 12 ". We conclude that jn;1 (a1 +    + an ) ; aj if
n  N.

FEJER KERNEL

31

Ex. 4.5 Give an example of an unbounded sequence which is Cesaro convergent.

Ex. 4.6 Which of the following statements are true?

(i) If (C) limn!1 an = a then (C) limn!1 (an )3 = a3 .

P
(ii) If (C) 1
n=1 an = s then (C) limn!1 an = 0.

4.2 Denition of Fejer kernel

4.7 In (3.2) we introduced the partial Fourier sums; Sn f ] of a
1function f 2 L12 . In

Chapter 3 we examined convergence of the sequence (SN f ])(x) N =0 for
xed x. More
generally ;we may examine
Cesaro convergence of this sequence, i.e., convergence of the

sequence ( N f ])(x) 1
,
where
N =0

N f ] := S0 f ] + S1 Nf ] ++ 1   + SN f ] (f 2 L12
N = 0
1
2
: : :):

(4:3)

The function N f ] is called the N th Cesaro mean of f .

4.8 Denition-Proposition The function in (4.3) can be written as

Z
Z
1
1
f (t) KN (x ; t) dt = 2
f (x + t) KN (t) dt
(4:4)
( N f ])(x) = (f  KN )(x) = 2
;
;

where KN (x) is the Fej er kernel:

N
N N + 1 ; jnj
X
X
inx
KN (x) := N 1+ 1 Dn (x) =
e
n=0
n=;N N + 1
8
sin( 21 (N + 1)x)
2
>
< 1
(x 2= 2Z),
sin( 21 x)
=> N + 1
:
N +1
(x 2 2Z).

(4:5)

Proof The second equality in (4.5) follows from the de
nition of Dn (x) in (3.4). From
(3.4) we get also that

KN (x) =

sin( 21 x) + sin( 32 x) +    sin((N + 12 )x)

:
(N + 1) sin( 21 x)

Now multiply numerator and denominator of this last expression by sin( 21 x) and use that
sin( 21 x) sin((k+ 12 )x) = 12 (cos(kx);cos((k+1)x)) and 1;cos((N +1)x) = 2 sin2 ( 21 (N +1)x):
This proves the last equality in (4.5). From (4.5) we see that KN 2 C2 and that it is an
even function. The third equality in (4.4) uses these last facts.

32

CHAPTER 4

The Fejer Kernels K8, K16, centered at 0
4 respectively.

4.3 Cesaro convergence of Fourier series

4.9 We need the following three easy consequences of (4.5):
KN (x)  0 (x 2 R )

1 Z K (t) dt = 1

2 ; N
1
2
KN (x)



(0 < jxj
):
(N + 1) sin2 ( 21 x) (N + 1) x2

(4:6)
(4:7)
(4:8)

4.10 Theorem (Fejer) Let f 2 L12 , a 2 R . Suppose that f is continuous in a. Then

lim ( f ])(a) = f (a):
N !1 N

Moreover, if f 2 C2 then the convergence in (4.9) is uniform for a 2 R .

(4:9)

FEJER KERNEL

33

Proof Let " > 0. We want to show that j N (a) ; f (a)j < " for N suciently large. It
follows from (4.4) and (4.7) that

Z
1
N (a) ; f (a) = 2 (f (a + t) ; f (a)) KN (t) dt:
;
R
R
R
For any  2 0
) we split up the above integral as ; = jtj< + <jtj< . Then, by use
of (4.6),
Z
Z
1
1
jf (a + t) ; f (a)j KN (t) dt:
j N (a) ; f (a)j
2 jf (a + t) ; f (a)j KN (t) dt + 2
;
<jtj<
Denote the two successive terms on the right hand side by I1 and I2. First we estimate
I1. By the continuity of f in a we can choose  such that jf (a + t) ; f (a)j
21 " if jtj
.
Keep this value of . Then, by use of (4.7), I1
12 ". Next we estimate I2 . It follows from
(4.8) that
Z
Z


;
2
I2
2(N + 1)
jf (a + t) ; f (a)j t dt
2(N + 1)2 (jf (a + t)j + jf (a)j) dt
<jtj<
;
2

(N + 1)2 (kf k1 + jf (a)j):
Hence I2
12 " for N suciently large. This proves the
rst part of the theorem.
Next assume that f 2 C2 . Then f is uniformly continuous on R (by the compactness
of ;
] and by the 2-periodicity). Thus we can choose  such that jf (a + t) ; f (a)j
12 "
for all a 2 R if jtj
. With this choice of  we have I1
21 " for all a. Then
2

I2
(N + 1)2 (kf k1 + kf k1 ):

Hence I2
12 " for all a if N is suciently large. We conclude that j N (a) ; f (a)j < " for
all a if N is suciently large.

4.11 Corollary (Fejer) Let f 2 L12 , let a 2 R , and suppose that the limits
f (a+ ) := lim
f (x)

x#a
exist. Then we have:

f (a;) := lim
f (x)
x"a

lim ( f ])(a) = 21 (f (a+) + f (a;)):

N !1 N

Ex. 4.12 Give the proof of Corollary 4.11 (compare with the proof of Theorem 3.5).
4.13 Theorem 1.11(a) is an immediate corollary of the second statement in Theorem
4.10:

34

CHAPTER 4

Corollary (Weierstrass) The space of trigonometric polynomials is dense in C2 with

respect to the norm k : k1 .
Proof Let f 2 C2 . Let " > 0. Then, by the uniform convergence stated in Theorem
4.10, we can take N such that k N f ];f k1 < ". Now observe that N f ] is a trigonometric
polynomial.

Let f 2 L12 and a 2 R. Suppose that f is continuous in a. If

limN !1 (SN f ])(a) exists then this limit is equal to f (a).
Proof By Proposition 4.4 the limit is equal to limN !1 ( N f ])(a) and hence, by Theorem
4.10 to f (a).

4.14 Corollary

4.15 Remark Observe that N f ] := f  KN (f 2 L12 ), where KN R(N = 0
1
2
: : :)

is an even nonnegative function belonging to L12 with properties (2);1 ; KN (t) dt = 1

and:

For each  2 (0
): Nlim

!1 KN (x) = 0 uniformly for 
jxj
 .
Observe that the proof of Theorem 4.10 only uses these properties of KN and N f ].
Hence Theorem 4.10 remains valid for any choice of the functions KN satisfying the above
properties.

Ex. 4.16 Let f 2 L12 . Show that limN !1 N f ] = f in L12 .

Hint Show that limN !1 k N f ] ; f k1 = 0 for f 2 C2 and use that C2 is dense in L12 .

4.4 Another approximation theorem of Weierstrass

Corollary 4.13 is an approximation theorem involving trigonometric polynomials. From
this we can derive an even more famous approximation theorem of Weierstrass involving
ordinary polynomials. (There exist many other proofs of this theorem.) As a tool for the
proof we introduce Chebyshev polynomials.

4.17 Denition-Proposition For each nonnegative integer n there is a unique polynomial Tn in one variable such that

Tn (cos t) = cos(nt) (t 2 R ):

(4:10)

The polynomial Tn has degree n. It is called a Chebyshev polynomial (of the
rst kind).
Proof For x 2 ;1
1] Tn (x) is uniquely determined by (4.10). Hence, if Tn is a polynomial then it is unique. Since
cos t cos nt = 12 cos(n + 1)t + 21 cos(n ; 1)t

we have

Tn+1 (x) = 2x Tn (x) ; Tn;1 (x) (n = 1
2
: : :)

while T0 (x) = 1, T1(x) = x. Now it follows by induction with respect to n that Tn (x) is a
polynomial of degree n in x.

FEJER KERNEL

35

Ex. 4.18 Show that

Z

( 0
n 6= m,
dx
= =2
n = m 6= 0,
Tn (x) Tm(x) p
1 ; x2
;1

n = m = 0.
1

Thus the Tn form an orthogonal system in L2((;1
1) (1 ; x2 ); 12 dx). Show also that this
orthogonal system is complete.

Ex. 4.19 Show that for each nonnegative integer n there is a unique polynomial Un (x)
of degree n in x such that

Show that

n + 1)t (0 6= t 2 R ):
Un (cos t) = sin(sin
t

(4:11)

Un+1 (x) = 2x Un (x) ; Un;1 (x) (n = 1
2
: : :)

while U0 (x) = 1, U1 (x) = 2x. The polynomial Un is called a Chebyshev polynomial (of
the second kind). Show that

Z1
p
m,
2
Un (x) Um(x) 1 ; x dx = 0=
2
nn 6=
=
m.
;1

Thus the Un form an orthogonal system in L2((;1
1) (1 ; x2 ) 21 dx). Show also that this
orthogonal system is complete.

4.20 Theorem (Weierstrass) Let f 2 C (a
b]). Then there exists for each " > 0 a
polynomial P in one variable such that jf (x) ; P (x)j < " for x 2 a
b].
Proof Without loss of generality we may assume that a
b] = ;1
1] (why?). Put g(t) :=
f (cos t). Then g 2 C2 and g is even, so gb(n) = bg(;n). Hence
N N +1;n
X
( N g])(t) = gb(0) + 2
gb(n) cos(nt)
n=1 N + 1

and N g] ! g for N ! 1, uniformly on R (see Theorem 4.9), so certainly uniformly on

0
]. Put
N
X
fN (x) := gb(0) + 2 N N+ +1 ;1 n bg(n) Tn(x):
n=1
Then fN (cos t) = ( N g])(t) and fN (x) is a polynomial of degree
N in x. Since N g] ! g
for N ! 1, uniformly on 0
], we have that fN ! f as N ! 1, uniformly on ;1
1].

Ex. 4.21 Let f (x) := jxj. Determine a sequence of polynomials which converges uniformly to f on ;1
1].
Ex. 4.22 The analogue of Theorem 4.20 does not hold on R . Give an example of a
continuous function on R which cannot be uniformly approximated by polynomials on R .

36

CHAPTER 4

4.5 Further exercises

Ex. 4.23 For 0
r < 1 let
1
X
Pr (x) :=
rjnj einx
n=;1

(x 2 R )

(4:12)

the so-called Poisson kernel. For f 2 L12 put

Ar f ] := f  Pr :

(4:13)

The functions Ar f ] are known as Abel means. Let a 2 R and suppose that f is continuous
in a. Prove that
(4:14)
lim
(A f ])(a) = f (a):
r "1 r

Moreover show that, if f 2 C2 , the convergence in (4.14) is uniform for a 2 R .

Hint Use Remark 4.15.

Ex. 4.24 Prove that the Gibbs phenomenon cannot occur for the Cesaro and Abel means.
That is, if m < f < M , then m < Ar f ]
CN f ] < M .

5 Some applications of Fourier series

37

5.1 The isoperimetric inequality

5.1 Theorem The area A of a region in the plane which is encircled by a closed nonsel
ntersecting C 1 -curve of length L satis
es A
L2 =(4). Equality holds i the curve is
a circle.

Proof Without loss of generality we may assume that L = 2, and that the curve is

positively oriented and parametrized by its arc length. We may also identify the plane
with C . Then the curve has the form t 7! f (t) with f 2 C21 and with jfR0(t)j = 1 for all
t. Furthermore we may assume without loss of generality that fb(0) = 21 ; f (t) dt = 0.
Then we have to show that A
 with equality i f (t) = ei(t+t0 ) for some t0 2 R . Now
we have
Z2
(2)
(1) 1
A = 2 Im
f 0 (t) f (t) dt =  Im hf 0
f i
 jhf 0
f ij
 kf 0 k2 kf k2
(3)

0
=  kf
 kf k2 (4)

(5)

= :
; fb(0)k2
 kf 0k2 (6)

(5:1)

Equality (1) follows from:

Ex. 5.2 Let t 7! f (t) be a positively oriented closed Rnon-sel
ntersecting C 1 -curve in C .

Show that the area of the encircled region equals 12 Im

2
0

f 0(t) f (t) dt.

Inequality (2) in (5.1) is the Cauchy-Schwarz inequality. Equalities (3) and (6) use that
kf 0k2 = 1 by the assumption jf 0(t)j = 1. Equality (4) uses the assumption fb(0) = 0.
Equality (5) follows from:

Ex. 5.3 Let f 2 C21 . Show that kf ; fb(0)k2
kf 0 k2 with equality i fb(n) = 0 for
n=
6 ;1
0
1.
The proof of the last part of Theorem 5.1 is also left as an exercise:

Ex. 5.4 Show that equality everywhere in (5.1) implies that f (t) = ei(t+t0 ) for some
t0 2 R .

5.2 The heat equation

5.5 The partial dierential equation
@ F (t
x) = 1 @ 2 F (t
x) ((t
x) 2 (0
1)
R )
2 @x2
@t
is called the heat equation. It is often considered with initial value
F (0
x) = f (x) (x 2 R )

(5:2)
(5:3)

38

CHAPTER 5

for a given function f on R. We will assume that f 2 C21 and we are interested in
a function F which is continuous on 0
1)
R and C 1 on (0
1)
R , which satis
es
F (t
x + 2) = F (t
x) for all t  0, and which is a solution of (5.2), (5.3). In a suitable
normalization of the variable t we can interprete this as the evolution in time of the
temperature on a ring which is idealized as the unit circle parametrized by its arc length.
Then F (t
x) is the temperature at position x and time t. The function f gives the initial
temperature distribution at time t = 0.
Put Ft (x) := F (t
x) and
Z
1
n (t) := Fbt (n) = 2
F (t
x) e;inx dx:
;
Ex. 5.6 Let f 2 C21 and let F be as above. Show the following.
(a) n is continuous on 0
1).
(b) n is C 1 on (0
1) with kth derivative given by
Z @ k F (t
x)
1
(
k
)
;inx dx:
n (t) = 2
e
k
@t
;
(c) We have
n0 (t) = ; 21 n2n (t)
n (0) = fb(n)
n (t) = fb(n) e; 12 n2 t :

Ex. 5.7 Let f 2 C21 . Show that

1
X
F (t
x) :=
fb(n) einx; 21 n2 t
n=;1

(5:4)

is a well-de
ned C 1 function on (0
1)

R , continuous on 0
1)
R , 2-periodic in x
and satisfying (5.2), (5.3). Conclude that F (t
x) given by (5.4) is the unique solution of
(5.2), (5.3) of this type.
Hint For the
rst part use Theorem 2.5(b).

Ex. 5.8 Let F be given by (5.4). Show that

Z

Ft (x) dx =

f (x) dx (t > 0)

(i.e., the total heat on the ring remains constant in time), and
lim F (t
x) = fb(0)
t!1

(i.e., the temperature distribution tends to the constant distribution as t ! 1).

Ex. 5.9 De
ne the heat kernel by

Ht (x) :=

X inx; 1 n2 t
e 2
(t > 0):

n2Z

Prove that Ht 2 C21 . Show that, for F given by (5.4), we have Ft = f  Ht (t > 0).

Ex. 5.10 Show that (f  Hs)  Ht = f  Hs+t (s
t > 0) and give a physical explanation
of this formula.

APPLICATIONS

39

5.3 Weyl's equidistribution theorem

5.11 Theorem Let 0 <  2 R n Q and let 0
a < b
1. Then
lim #fr 2 N j 0 < r < n
< r > 2 a
b]g = b ; a:

n
Here # denotes the number of elements of a set and < x > denotes the fraction al
part of a positive number x.
n!1

Ex. 5.12 Prove this theorem by completing the following outline:

1) Let f 2 C2 and  2 R n Q . Prove that
lim

n!1

Pn

j =1 f (2j ) = 1

2

2
0

f (t) dt:

In case f  1.
In case f (t) = eikt , k 2 Z.
In case f is a goniometric polynomial.
For general f 2 C2 .
2) Prove that the formula is also valid if f is the periodic extension of the characteristic
function ab] of a closed interval
a
b]  (0
2). (Look for continuous functions f1
f2
R2
such that f1
f
f2 and 0 (f2 ; f1)(t) dt < ) and apply 1).
Derive the equidistribution theorem.
a.
b.
c.
d.

40

6 Generalities about Fourier integrals

This chapter is the beginning of Part II of the Syllabus. It deals with Fourier integrals.

6.1 In Part I we have seen that, for a 2-periodic function f which is suciently smooth,
for instance f 2 C22 , we can write

Z
1
;
int
ina
f (a) =
f
(
t
)
e
dt
e
(a 2 R )

(6:1)
n=;1 2 ;
where the doubly in
nite series converges absolutely. Now we will consider functions f on
R which are usually not periodic. The analogue of (6.1), valid for suciently nice functions
f , is


Z1 Z1
1
;
itx
f (a) = 2
f (t) e dt eixa dx (a 2 R ):
(6:2)
;1 ;1
If f is a function on R which is suciently smooth (say C 2 ) and which decreases suciently
fast in absolute value to 0 as t ! 1 (say f (t) = O(jtj;2) as jtj ! 1), then (6.2) holds
with absolutely convergent inner and outer integrals. This we will prove later. For the
moment we will only prove (6.2) heuristically from (6.1).
1
X

6.2 Let f be a C 2 -function on R with compact support and let a 2 R . We will give a
heuristic proof of (6.2). For  > 0 such that supp(f )  ;
] and jaj
 de
ne a
function f 2 C22 such that f (t) := f (t) if ;
t
. It follows from (6.1) that


1 Z
X
1
;
int
f (a) = f (a=) = 2
f (t) e dt eina=
n=;1 ;


Z
1
1
X
1
;
int=
f (t) e
dt eina= ;1
= 2
n=;1 ;1


X Z1
1
;
ixt
= 2
f (t) e dt eixa ;1
x2;1 Z ;1
X
= 21 Nlim
fb(x) eixa ;1

!1 x2;1 Z\;NN ]

where we have put

fb(x) :=

Z1
;1

f (t) e;ixt dt

(x 2 R ):

(6:3)
(6:4)

The sum in (6.3) can be considered as a Riemann sum for the partition of ;N
N ] into
equidistant points with distance ;1 . Since the function x 7! fb(x) eixa is continuous, this
Riemann sum tends to the corresponding integral. So formally the limit of (6.3) as  ! 1
equals
1 lim Z N fb(x) eixa dx

2 N !1 ;N
which is equal to the right hand side of (6.2). It has to be justi
ed that we may take the
limit for  ! 1 in (6.3) inside the limit for N ! 1. We will skip this here.

GENERALITIES

41

6.3 Denition Let f 2 L1(R ). Then the integral on the right hand side of (6.4) is
absolutely convergent since jf (t) e;ixtj
jf (t)j. So (6.4) yields a well-de
ned function fb
on R . We call the transform f 7! fb given by (6.4) the Fourier transform. The function fb
is called the Fourier transform of f . Note that fb only depends on the equivalence class of
f . So the function fb is well-de
ned for f 2 L1 (R ).

Ex. 6.4 Let a
b] be a closed bounded interval and let f := ab] be its characteristic
function. Show the following: f 2 L1 (R) and
8 ;iax ;ibx
< e ; e = 2 sin( 12 (b ; a)x) e; 12 i(a+b)x if x 6= 0,
fb(x) = :
ix
x
b;a
if x = 0.

Observe from this result that fb is continuous and that limx!1 fb(x) = 0:

Ex. 6.5 Let f (x) = 1+1x2 on R . Compute fb. (Use Contour integration.)
6.6 Let p 2 1
1), in particular we will consider p = 1 or 2. We write

kf kp :=

Z1
;1

1=p

jf (t)jp dt

if f 2 Lp (R) or f 2 Lp (R )

and we put

kf k1 := sup jf (x)j if f is a bounded continuous function on R :

x2R

Let Cc (R ) denote the space of complex-valued continuous functions on R with compact

support. The following Proposition is proved in Rudin, Theorem 3.14.

6.7 Proposition Let p 2 1
1). Then Cc (R ) is dense in Lp (R ).

Ex. 6.8 A function f : R ! C is called simple if it is a linear combination of characteristic

functions of bounded intervals. Equivalently, f is simple if there are real numbers a1 <
a2 < : : : < an such that f vanishes outside a1
an] and f is constant on each interval
(ai
ai+1) (i = 1
: : :
n ; 1).
Let p 2 1
1). Prove, by using Proposition 6.7, that the space of simple functions is dense
in Lp (R ).

6.9 Theorem Let f 2 L1(R ). Then

(a)
(b)
(c)

fb is a bounded continuous function on R .

kfbk1
kf k1.
limx!1 fb(x) = 0 (Riemann-Lebesgue Lemma).

42

CHAPTER 6

Proof For the proof of the continuity of fb note that, for x 2 R ,

Z1

lim

y!x ;1

f (t) e;iyt dt =

Z 1

Z1

;
iyt
lim
f (t) e
dt =
f (t) e;ixt dt
y
!
x
;1
;1

;iyt
by Lebesgue's
R 1 dominated convergence theorem (syll. Integratietheorie, 2.15) since jf (t) e j

jf (t)j and ;1 jf (t)j dt < 1. Clearly, it follows from (6.4) that
Z1
b
jf (x)j

jf (t)j dt = kf k1 (x 2 R ):
;1

This settles (a) and (b).

For the proof of (c)
rst note that the result is valid if f is a simple function. This
follows from Exercise 6.4. Now let f 2 L1(R ). Let " > 0. By Exercise 6.8 there is a simple
function g such that kf ; gk1 < "=2. By (b) we have that jfb(x) ; gb(x)j < "=2. Hence
jfb(x)j
"=2 + jgb(x)j and there is M > 0 such that jgb(x)j < "=2 if jxj > M .

Ex. 6.10 Let

C0 (R) := ff : R ! C j f is continuous and x!1
lim f (x) = 0g:
Prove that C0 (R ) is a closed linear subspace of the Banach space of bounded continuous
functions on R with respect to the sup-norm, so C0 (R ) is a Banach space itself with respect
to the sup-norm. Also observe that the map f 7! fb: L1(R ) ! C0 (R ) is a bounded linear
operator.

Ex. 6.11 Let f 2 L1(R ), a 2 R , c 2 R nf0g. Prove the following.

(a)
(b)
(c)
(d)

If g(t) := f (t + a) then gb(x) = eiax fb(x).

If g(t) := eiat f (t) then bg(x) = fb(x ; a).
If g(t) := f (ct) then gb(x) = jcj;1 fb(c;1 x).
In particular, if g(t) := f (;t) then gb(t) = fb(;t).

6.12 Proposition Let f
g 2 L1(R ). Then

Z1

;1

f (x) bg(x) dx =

Z1
;1

fb(x) g(x) dx:

(Both integrals converge absolutely because of Theorem 6.9(a).)

Proof We have to show that


Z1
Z1
;
ixy
f (x) e g(y) dy dx
x=;1

is equal to

Z1
y=;1

y=;1

Z1
x=;1

f (x) e;ixy g(y) dx dy:

GENERALITIES

43

This follows from the Fubini Theorem 2.14 since

Z1
Z1
;
ixy
jf (x) e g(y)j dy dx < 1:
x=;1

y=;1

6.13 Proposition Let f 2 C 1 (R ) such that f
f 0 2 L1(R ). Then

(f 0)b(x) = ix fb(x)

(6:5)

and fb(x) = o(jxj;1 ) as x ! 1.

Proof Let M > 0. Integration by parts shows that
ZM
ZM
0
;
ixt
;
ixM
ixM
f (t) e dt = e
f (M ) ; e f (;M ) + ix
f (t) e;ixt dt:
;M

;M

Then (6.5) will follow by letting M ! 1 in the above formula, provided we can show
that limRM !1 f (M ) = 0. These last limits are indeed valid.R For instance, f (M ) =
1 0
f (0) + 0M f 0(t) dt. Hence f (1) := lim
R 1M !1 f (M ) = f (0) + 0 f (t) dt exists. Then
necessarily f (1) = 0, since otherwise ;1 jf (t)j dt = 1.
The second statement follows by combination with Theorem 6.9(c).

6.14 Proposition Let f 2 L1(R ) such that the function g: x 7! xf (x) is in L1 (R ).

Then fb is dierentiable on R with ( fb)0 = ;igb.
Proof Let a
b 2 R with a < b. Then, by Proposition 6.12 and Exercise 1.4,
Zb
a

gb(x) dx =
=
=

Z1

Z;1
1
Z;1
1

=i

bg(x) ab](x) dx
g(x) (ab])b(x) dx
xf (x) e

Z;11

;1

;iax ; e;ibx

ix

dx

f (x)(e;ibx ; e;iax ) dx = i(fb(b) ; fb(a)):

Hence

fb(a + h) ; fb(a) = ;ih;1 Z a+h bg(x) dx:

h
a
Now let h ! 0 and use the fact that gb is continuous.

6.15 Corollary
(a) Let f 2 C k (R) such that f
f 0
: : :
f (k) 2 L1(R ). Then (f (k) )b(x) = (ix)k fb(x) and
fb(x) = o(jxj;k ) as x ! 1.
R1
(b) Let f be a Lebesgue measurable function on R such that ;1
(1 + jtj)k jf (t)j dt < 1.
R
1 (;it)k f (t) e;ixt dt.
Then fb 2 C k (R ) and ( fb)(k) (x) = ;1
(c) Let 2 R with > 1. Let f be a Lebesgue measurable function on R such that
f (x) = O(jxj;) for jxj ! 1. Then fb 2 C k (R ) if k + 1 < .

44

CHAPTER 6

Observe the general rules:

 Faster decrease of jf (t)j to 0 as jtj ! 1 implies more smoothness of fb.
 More smoothness of f implies faster decrease of jfb(x)j to 0 as jxj ! 1.

6.16 We introduce the following notation for some operators acting on functions on R :
(Df )(x) := f 0(x) (di
erentiation)

(Mf )(x) := x f (x) (multiplication)

(Ta f )(x) := f (x + a) (translation)

(Ea f )(x) := eiax f (x) (exponential multiplication)

(Pf )(x) := f (;x) (parity operator):

Some of the previous results can now be summarized in the following table:

g
Df
Mf
Ea f
Ta f
t 7! f (ct)
Pf

gb
iM fb
iDfb
T;a fb
Ea fb
x 7! jcj;1 fb(c;1 x)
P fb

6.17 Denition S is the set of all f 2 C 1 (R ) such that, for all nonnegative integers
n
m the function x 7! xn f (m) (x) is bounded.
So S consists of the C 1 -functions f on R such that f and all its derivatives are
rapidly decreasing to 0 as the argument tends to 1. Brie"y we call such functions

rapidly decreasing C 1 -functions on R . This class of functions was introduced by Laurent

Schwartz. He used this as a class of test functions for his so-called tempered distributions.
The (immediate) proofs of the following statements are left to the reader.

6.18 Proposition S is a linear space. If f 2 S then the functions Df , Mf , Ea f , Ta f ,

x 7! f (cx) and Pf are also in S . If f 2 S then fb 2 S .
Ex. 6.19 Let f (t) := e; 21 t2 (the Gaussian function). Show the following:
(a) f 2 S .
(b) Df = ;Mf , Dfb = ;M fb.
p
R 1 ; 21 x2
e
dx = 2.
(c) fb(x) = C e; 12 x2 with C = ;1
Hint Solve the dierential equation for fb in (b).

Ex. 6.20 Let f (t) := e; 12 t2 . Then

fb(x) =

Z1 1
Z1 12
1 x2
;
t
;
ixt
;
2
2
e e dt = e
e; 2 (t+ix)2 dt (x 2 R ):
;1

;1

Derive from this formula the result of Exercise 6.19(c), by contour integration.

GENERALITIES

45

Ex. 6.21 Let f 2 L1(R ). Suppose that f has compact support, i.e., f vanishes outside

some bounded interval. Prove the following.

(a) If fb = 0 then f = 0 a.e..
Hint Reduce this to the case of periodic functions.
(b) The function fb is the restriction to R of an entire analytic function on C (the function
given by (6.4) for x 2 C ).
(c) If fb also vanishes on R outside some bounded interval then f = 0 a.e..

6.22 Denition-Theorem For f
g 2 L1(R ) the convolution product f  g is de
ned

by

(f  g)(x) :=

Z1

;1

f (y) g(x ; y) dy

(x 2 R ):

(6:6)

Then the integral in (6.6) converges absolutely for almost all x 2 R and f  g 2 L1 (R ).
Furthermore,
kf  gk1
kf k1 kgk1
(6:7)
and
(6:8)
(f  g)b(x) = fb(x) bg(x):
Proof Analogous to the proof of De
nition-Theorem 2.19. Use the Fubini Theorem 2.14.

Ex. 6.23 Let p(x) be a polynomial with real coecients in the real variable x. Find
necessary and sucient conditions in order that the function x 7! ep(x) belongs to S .

46

7 Inversion formula

7.1 (Dirichlet kernel)

Let f 2 L1 (R), ! > 0. Put

Z!
1
(I! f ])(s) := 2
fb(x) eixs dx (s 2 R ):
;!

(7:1)

In view of (6.2) we expect (but still have to prove) that I! f ] ! f as ! ! 1. The

advantage of (7.1) compared to the corresponding integral from ;1 to 1 is that it can be
written as an integral transformation (of convolution type) acting on f with very simple
integral kernel: the Dirichlet kernel
sin(!t) = 1 Z ! e;ixt dx:
(7:2)
t
2 ;!
Proposition We have
Z1
(I! f ])(s) =
f (t + s) sin(t!t) dt

(7:3)
;1


sin(
!t
)


(7:4)
t 
!= (t 2 R )

Z 1 sin t
Z 1 sin(!t)
2
dt = 
(7:5)
t dt = 1
0
;1 t
(the last two integrals converging but not absolutely converging).
The proof of these formulas is left to the reader. For the proof of (7.3) substitute (6.4)
into (7.1) and interchange the two integrals (allowed by Fubini's theorem check this).

7.2 The following results are analogous to x3.2. We give a convergence criterium for
Fourier inversion in a certain point.

Theorem Let f 2 L1 (R), a 2 R . Suppose that one of the following two conditions holds:
(a) There are M

 > 0 such that

jf (a + x) ; f (a)j < M jxj for jxj < .

(b) f is continuous in a and also right and left dierentiable in a.
Then we have:
lim (I f ])(a) = f (a):
!!1 !

(7:6)

Proof Condition (b) implies condition (a), so we may assume Condition (a). We have
Z 1 sin(!x)
dx
(I! f ])(a) ; f (a)
;1 x
Z 1 f (a + x) ; f (a)
Z
f (a + x) sin(!x) dx:
=
sin(
!x
)
dx
+
x
;1
jxj>1 x

(7:7)

INVERSION FORMULA

47

Since

Z ! sin y
Z 1 sin y
Z 1 sin(!x)
2
dx =
dy ;! 
y dy = 1 as ! ! 1,
;! y
0
;1 x
it is sucient to prove that the right hand sideRtends
to 0 as ! ! 1. However, we will
1
;
1
see that the right hand side is of the form  ;1 g(x) sin(!x) dx with g 2 L1(R ), so it
tends to 0 as ! ! 1 because of the Riemann-Lebesgue lemma (see Theorem 6.9(c)).
The claim about the right hand side is proved by splitting it up as a sum of three
terms (where we assume that  < 1):


f
(
a
+
x
)
;
f
(
a
)
 < M jxj;1

jxj < : jg(x)j = 
x


f
(
a
+
x
)
;
f
(
a
)

;1 (jf (a + x)j + jf (a)j)



jxj < 1: jg(x)j = 
x


f
(
a
+
x
)
jxj  1: jg(x)j =  x 
jf (a + x)j:

7.3 Theorem Let f 2 L1(R ), let a 2 R , and suppose that the limits
f (a+ ) := lim
f (x)

x#a

f (a;) := lim
f (x)
x"a

exist. Suppose that one of the two following conditions holds:

(a) There are M

 > 0 such that

jf (a + x) ; f (a+)j
Mx and jf (a ; x) ; f (a;)j
Mx for 0 < x < :

(b) f is right and left dierentiable in a in the sense that the following two limits exist:

f (a + y) ; f (a+)

f 0 (a+) := lim
y#0
y
Then we have:

f (a ; y) ; f (a;) :
f 0 (a;) := lim
y #0
;y

lim (I f ])(a) = 21 (f (a+) + f (a; )):

!!1 !

Ex. 7.4 Prove Theorem 7.3 as a corollary of Theorem 7.2, analogous to the proof of
Theorem 3.5.

Let f
g 2 L1(R ), let a 2 R , and suppose

that f (x) = g(x) for x in a certain neighbourhood of a. Then precisely one of the following
two alternatives holds for the behaviour of (I! f ])(a) and (I! g])(a) as ! ! 1:
(a) Both limits exist and are equal:

7.5 Corollary (localization principle)

lim (I f ])(a) = !lim

!!1 !
!1(I! g ])(a)
(b) Neither (I! f ])(a) nor (I! g])(a) has a limit as ! ! 1.

48

CHAPTER 7

Thus, if for some f 2 L1 (R) and some a 2 R the inversion formula (7.6) holds, then
this inversion formula remains valid if f is perturbed within L1(R ) such that f remains
unchanged on some neighbourhood of a.

Ex. 7.6 Prove Corollary 7.5.

7.7 Corollary Let f 2 L1 (R) such that also fb 2 L1(R ) and f is dierentiable on R .

Then (fb )b is well-de
ned and

(fb )b(s) = 2 f (;s) (s 2 R ):
Ex. 7.8 Prove Corollary 7.7. Can the dierentiability assumption be relaxed?
De
ne f : R ! R by
1,
f (x) := 10 ; jxj ifif jjxxjj

> 1.
(a) Show that f 2 L1 (R) and fb(x) = 2x;2 (1 ; cos x) (x 6= 0), fb(0) = 1.
(b) Show that fb 2 L1 (R) and that (fb )b(a) = 2 f (;a) for all a 2 R .
(c) Show that
Z 1 1 ; cos x
dx = :
;1 x2
Ex. 7.10 Let
81
< 2 ; t if 0 < t
12 ,
f (t) := : ; 21 ; t if ; 12
t < 0,
0

otherwise.

Ex. 7.9 (van Rooij, Voorbeeld 15.8)

Compute fb(x). Examine the behaviour of fb(x) as jxj ! 1.

Ex. 7.11 Let f be continuously dierentiable on R nf0g. Suppose that f
f 0 2 L1(R ) and
that the following four limits exist:

f (t) ; a
lim f (t) ; b :

a := lim
f
(
t
)


b
:=
lim
f
(
t
)


lim
t"0
t#0
t"0
t#0
t
t

(a) Show that fb(x) = O(jxj;1) as jxj ! 1.

(b) Show that fb(x) = o(jxj;1) as jxj ! 1 i a = b.

Ex. 7.12 Let

; t2 ); 12 if jtj < 1,

f (t) := (1
0
if jtj  1.
Assume that Re > ; 12 . Show that
1 (;x2 =4)k
;( 12 ) ;( + 21 ) X
b
f (x) = ;( + 1)


k=0 k! ( + 1)k

where ( + 1)k := ( + 1)( + 2) : : : ( + k).

What can be concluded about the speed with which jfb(x)j tends to 0 as jxj ! 1?

49

8 L2 theory

We will work now with the space L2(R ) of square integrable functions on R and with the
Hilbert space L2(R ) of equivalence classes of such functions. Inner product and norm are
given by

21
Z1
Z1
1
2
< f
g >:=
f (x) g(x) dx
kf k2 := (hf
f i) 2 =
jf (x)j dx : (8:1)
;1

;1

The space S (see De
nition 6.17) is a linear subspace of L2 (R ).

We will now denote by F a slightly renormalized version of the Fourier transform
f 7! fb:
Z1
1
1
(F f )(x) := p fb(x) = p
f (t) e;ixt dt:
(8:2)
2
2 ;1
Here f 2 L1(R ), in particular we will take f 2 S .

8.1 Proposition The map F : S ! S is a linear bijection, with inverse

(F ;1g)(s) = (F g)(;s) (g 2 S ):
Furthermore F preserves the inner product (8.1) on S :
hF f
F gi = hf
gi (f
g 2 S ):

(8:3)
(8:4)

For convenience we give also more extended versions of (8.3) and (8.4):
Z1
1
;
1
(F g)(s) = p
g(x) eisx dx

(8:5)
2 ;1
Z1
Z1
Z1
1
f (t) g(t) dt =
(F f )(x) (F g)(x) dx = 2
(8:6)
fb(x) gb(x) dx:
;1
;1
;1
The proof of this Proposition is immediate, in view of Corollary 7.7, Proposition 6.18
and Proposition 6.12.

8.2 We will now introduce some special C 1 functions, which we will need as tools (see
also van Rooij, x18.1).
Let a < b and put

8


< exp 1 ; 1
f (x) := :
x ; b x ; a if a < x < b,
0

otherwise.
Then f 2 C 1 (R ) and f (x)  0 for all x 2 R .
Let
Z x
Z b

F (x) :=
f (t) dt = f (t) dt :
a

Then F 2 C 1 (R ), F (x) = 0 if x
a, F (x) = 1 if x  b, and F is strictly increasing on

a
b].

50

CHAPTER 8

Ex. 8.3 Let a
b] be a closed bounded interval and ab] its characteristic function. Show
that ab] can be approximated arbitrarily closely in L2 norm by functions in S .
8.4 Proposition The space S is dense in L2 (R).
Proof By Exercise 6.8 the simple functions are dense in L2 (R). Next use Exercise 8.3.
8.5 As a corollary of the previous propositions we obtain:
Theorem (Parseval-Plancherel) The map F : S ! S has a unique extension to a
continuous linear map F : L2(R ) ! L2(R ). This extension is an isometry of Hilbert spaces:
hF f
F gi = hf
gi (f
g 2 L2 (R))

it is bijective, and it satis
es
(F 2f )(s) = f (;s) a:e:
We have obtained two de
nitions of the Fourier transform, which we will denote by
F1 resp. F2 for the moment:
Z1
1
f (t) e;ixt dt (f 2 L1(R ))

(F1f )(x) := p
2 ;1
2
(F2f )(x) :=(L2) nlim
!1 F1 fn
where Fn 2 S and (L ) nlim
!1 fn = f .
For f 2 L1 (R) \ L2(R ) it will turn out that F1 f = F2f .

Lemma Let f 2 L1(R ) \ L2(R ). Then there is a sequence (fn )1n=1 in S such that
kf ; fn k1 ! 0 and kf ; fn k2 ! 0 as n ! 1.
Proof Let " > 0. There exists M > 1 such that kf;MM ] ; f kpp< " for p = 1
2. Write
g := f;MM ]. Then there exists h 2 S such that kg ; hk2 < "=(2 M ) < ". We can even
choose h in this way such that it has support within ;2M
2M ] (why?). Thus
Z 2M
p
kg ; hk1 =
jg(x) ; h(x)j dx
2 M kg ; hk2 < ":
;2M

Hence kf ; hk1 < 2" and kf ; hk2 < 2".

8.6 Theorem If f 2 L1(R ) \ L2 (R ) then F1f = F2f .

Proof Take a sequence (fn )1n=1 in S such that
p kf ;fnk1 ! 0 and kf ;fn k2 ! 0 if n ! 1.
2
Then F2f = limn!1 F1fn in L (R ). Since 2 kF1 f ; F1fn k1
kf ; fn k1 , it follows
that (F1f )(x) = limn!1 (F1fn )(x) pointwise. From integration theory we know that a

convergent sequence in L2 (R ) always has a subsequence which converges almost everywhere

1
to the limit function. Hence the sequence (F1fn )1
n=1 has a subsequence (F1 fnk )k=1 such
that
(F2f )(x) = klim
(F f )(x) = (F1f )(x) a:e:
!1 1 nk

L2 THEORY

51

8.7 Theorem If f 2 L2(R ) then

(F f )(x) = (L2)

ZM
1
;ixt dt

p
f
(
t
)
e
lim
M !1 2 ;M

which can be rewritten as

Z 1
ZM
2
1

;
ixt
lim
(F f )(x) ; p
f (t) e dt dx = 0:
M !1 ;1 
2 ;M

Proof kf ; f;MM ] k2 ! 0 if M ! 1. Hence kF2f ;F2 (f;MM ])k2 ! 0 as M ! 1.

Because f;MM ] 2 L1 (R) \ L2(R ), it follows that F2 (f;MM ]) = F1 (f;MM ]).
Ex. 8.8 (van Rooij, Opgave 19.A) Show that not L1 (R)  L2(R ). But show that all
bounded functions in L1(R ) lie in L2(R ).
Ex. 8.9 (van Rooij, Voorbeeld 19.9)
7.9 in order to show that

Use the Fourier transform computed in Exercise

Z 1 sin t
4
2 :
dt
=
3
;1 t

Ex. 8.10 (van Rooij, Opgave 19.B) Let f
g 2 L2 (R). Then fg 2 L1 (R), thus fcg
exists. Take representatives in L2(R ) of f
b bg 2 L2(R ) and denote these representatives also
by f
b bg. Show that

2 fcg(x) =

Z1

;1

fb(t) bg(x ; t) dt:

Ex. 8.11 (van Rooij, Opgave 19.E) For n a nonnegative integer de
ne the function
hn 2 S and the Hermite polynomial Hn by
hn (x) := (x ; d=dx)n e;x2 =2

Hn (x) := ex2 =2 hn (x):

Prove the following.

(a) We have

Hn+1 (x) = 2x Hn (x) ; Hn0 (x) and Hn (x) = (;1)n ex2 (d=dx)n e;x2 :
Hn is a polynomial of degree n with real coecients the term of degree n in Hn has
coecient 2n .
p
(b) hcn = (;i)n 2 hn . (Thus hn is an eigenfunction of the Fourier transform L2(R ) !
L2 (R).)
(c) If n 6= m then hhn
hm i = 0. (So the functions hn form an orthogonal system in
L2 (R).)
Hint Let n > m. Substitute hn (x) = (;1)n ex2 =2 (d=dx)n e;x2
hm (x) = e;x2 =2 Hm(x),
and perform integration by parts.

52

CHAPTER 8

(d) If f 2 L2 (R ) and hf
hni = 0 for all n, then f (x) = 0 a.e.. (So the functions hn form
a complete orthogonal system in L2(R ).)R
1 f (x) e;x2 =2 P (x) dx = 0. By the domiHint For each polynomial P we have ;1
nated convergence theorem this implies that
Z1
f (x) e;x2=2 e;ixy dx = 0:
;1

Ex. 8.12 (For this exercise use results from both Fourier series and Fourier integrals.)
Below de
ne x;1 sin x for x = 0 by continuity.
(a) Let t 2 R . Show that for each x 2 (;
) we have

1 sin( (t ; n))
X
inx = eixt
e
n=;1  (t ; n)

with pointwise convergence. What is the evaluation of the sum on the left hand side
for other real values of x?
(b) Show that, for all n
m 2 Z, we have
Z 1 sin((t ; n)) sin((t ; m))
(t ; m) dt = nm

;1  (t ; n)
where the integral converges absolutely.
(c) Does there exist f 2 L2(R ) with f 6= 0 such that
Z1
t ; n)) dt = 0 for all n 2 Z ?
f (t) sin((t(;
n)
;1
(d) Let f 2 L2(;
]). Extend fb, de
ned as a function on Z by (1.8), to a function on
R by
Z
1
f (x) e;ixt dx (t 2 R ):
fb(t) := 2
Show that

1
X
t ; n))
b
f (t) =
fb(n) sin((t(;
n)
n=;1

with absolutely convergent sum.

(t 2 R )

53

9 Poisson summation formula

The Poisson summation formula is an equality of two doubly in
nite sums over sets of
equidistant points in R . The
rst sum involves a function f on R , the second one its
Fourier transform fb.

9.1 Theorem Let f 2 L1 (R) \ C (R ) such that f (t) = O(jtj;2) as jtj ! 1 and fb(x) =
O(jxj;2) as jxj ! 1. (Take for instance f 2 S .) Then
1
X
1
fb(k) eikx
f (x + 2l) = 2
k=;1
l=;1
1
X

(x 2 R )

(9:1)

where the series on both sides pointwise and absolutely converge to a function in C2 (i.e.,
to a 2-periodic continuous function).
Proof Put

gN (x) :=

N
X
l=;N

f (x + 2l)

g(x) := Nlim
g (x) =
!1 N

1
X
l=;1

f (x + 2l):

The last series converges pointwise and absolutely on R because f (x) = O(jxj;2) as jxj !
1. This series even converges uniformly for x in a compact subset of R . Hence g is
continuous on R . Clearly, g is also 2-periodic. Next observe that
!
Z X
Z1
1
1
X
jgN (x)j

jf (x + 2l)j and
jf (x + 2l)j dx =
jf (y)j dy < 1:
;

l=;1

;1

l=;1

This justi
es an application of the dominated convergence theorem as follows (k 2 Z):

Z
Z
;
ikx
g(x) e dx = Nlim
gN (x) e;ikx dx
!1
;
;
Z (2N +1)
Z1
;
iky
= Nlim
f (y) e dy =
f (y) e;iky dy = fb(k):
!1
;(2N +1)

;1

Hence, because fb(k) = O(jkj;2) as jkj ! 1, we conclude that g(x) equals the right hand
side of (9.1).
Specialization of (9.1) yields
1
X
l=;1

1
X
fb(k):
f (2l) = 21
k=;1

Here f is as in Theorem 9.1 and there is absolute convergence on both sides.

(9:2)

54

CHAPTER 9

Ex. 9.2 Let f be as in Theorem 9.1. Assume c > 0. Prove the following identities.
(a)
(b)
(c)
(d)
(e)

1
X
l=;1

1
X

l=;1

1
1 X
f (x + 2c l) = 2c
fb(c;1 k) eikc;1 x

1
f (2c l) = 2c

k=;1

1
X

k=;1

fb(c;1 k)

1
X
1
p
e; 12 c;2 k2 eikc;1 x

c 2 k=;1
l=;1
1
1
X
X
e; (x+c l)2 = c;1
e; c;2k2 e2 ikc;1x

1
X

e; 12 (x+2 c l)2 =

l=;1

1
X

l=;1

k=;1

e; c2 l2 = c;1

1
X

k=;1

e; c;2k2 :

Ex. 9.3 Show that (d) and (e) of Exercise 9.2 remain valid if c
x 2 C with j arg cj < .
One of the theta functions is de
ned by
1
X
#3(z j  ) :=
ei n2 e2inz
n=;1

(Im  > 0
z 2 C ):

Use (d) of Exercise 9.2 in order to show that

;

 ; 21 exp ;z2 =( ) #3 ( ;1z j  ;1 i) = #3(z j i )

Ex. 9.4 (van Rooij, Opgave 14.A)

(Re  > 0
z 2 C ):

Show that

1 =  cosh 

2
2
sinh 

n2Z n +

(
> 0):

Hint Apply (9.2) with f (x) := e;jxj.

Ex. 9.5 Apply (9.2) with f as in Exercise 7.10 in order to show that
1
X

1 = 2
2 (sin x)2
m=;1 (x + m)

(x 2 R nZ):

10 Some applications of Fourier integrals

55

10.1 Heisenberg's inequality

10.1 Lemma Let V be a complex vector space with hermitian inner product h :
: i and
with linear operators A
B : V ! V such that hAf
gi = ;hf
Agi and hBf
gi = ;hf
Bgi
for all f
g 2 V . Then
 1

h 2i (AB ; BA)f
f i
kAf k kBf k (f 2 V ):
(10:1)
For given f 2 V equality holds in (10.1) i Af = iBf for some  2 R or Bf = 0.
Proof Observe that
;

h 21i (AB ; BA)f
f i = ;2i1 hBf
Af i ; hAf
Bf i = Im hAf
Bf i

and

 ;


 
Im hBf
Af i 
hAf
Bf i
kAf k kBf k:
10.2 We now discuss a special case of the above lemma. Take V := S , Af := f 0,
(Bf )(x) := ixf (x). Then

1;
1
2i (AB ; BA)f (x) = 2 f (x):
Hence

21 1 Z 1

12
Z1
2
2
2
2
1
1
x jf (x)j dx
(f 2 S ): (10:2)
y jfb(y)j dy
2 kf k
kfbk2 ;1
;1
2
For given f 2 S equality
holds in (10.2) i f 0(x) = ; x f (x) for some  2 R , i.e., i
2
1
f (x) = const: e; 2 x for some  > 0. (For 
0 the function f is not in S .)
As a corollary of (10.2) we
nd that for f 2 S , x0
y0 2 R the following inequality
holds:

21

12 1 Z 1
Z1
1
2
2
2
2
1
(y ; y0) jfb(y)j dy : (10:3)
(x ; x0) jf (x)j dx
2
kf k
b
k
f
k
;1
;1
2
2
For
xed f 2 S the right hand side of (10.3) attains its absolute minimum for
R1 b 2
R1
x jf (x)j2 dx
;1 y jf (y )j dy :
;1
b


y
=

(
f
)
:=
x0 = (f ) := R 1
R
0
1
;1 jf (x)j2 dx
;1 jfb(y )j2 dy
Here (f ) is;Rthe mean value
of the stochastic variable x with respect to the probability
1 jf (x)j2 dx ;1 jf (x)j2 dx on R and (fb) is the mean value of the stochastic
distribution ;1
;R 1 b 2
;1 b 2
variable y with respect to the probability distribution ;1
jf (y)j dy jf (y)j dy on R .
Denote the corresponding variances by

12
Z 1;

2
1
2
(f ) := kf k
x ; (f ) jf (x)j dx

;1
2

21
Z 1;

1
2
y ; (fb) jfb(y)j2 dy :
(fb) := b
kf k2 ;1

56

CHAPTER 10

Then we
nally obtain Heisenberg's inequality

(10:4)
(f ) (fb)  21 :
Thus the two stochastic variables x and y cannot be concentrated both on arbitrarily small
intervals.
Heisenberg's inequality is better known as Heisenberg's uncertainty relation when interpreted in quantum mechanics or in signal analysis. In quantum mechanics the variables
x and y are the position and impulse operator, respectively, for a one-dimensional particle.
Then position and impulse cannot be both measured with arbitrary accuracy. In signal
analysis x is the time and y is the frequency of a signal. Here time and frequency cannot
both be measured with arbitrary accuracy.

Ex. 10.3 Let f 2 S , kf k2 = 1,

a
b > 0, ( + a2)(
+ b2 ) < 41 . Show that it is

impossible that the following two inequalities hold together:

Z
Z
2
2
x jf (x)j dx < and
y2 jfb(y)j2 < 2
:
jxj>a

jyj>b

10.2 Fourier Transform on nite cyclic groups

The impact of the seemingly pure mathematical topic of this section on numerical and
applied mathematics cannot be overestimated. In the present section all relevant groups
will be
nite. We can endow such group with the discrete topology (all subsets are open)
and obtain what is called a topological group.
For completeness, we mention here that a topological group is a group endowed with
a topology such that the group operations multiplication and inversion are continuous. In
particular the notion of a continuous function on a topological group is available. Important
examples of topological groups are R n and C n with the Euclidean topology. We will not
pursue the implications of this notion at all, but we will use the terminology in the following
sense. The de
nitions will always be for a topological group G, in the results the group
will always be
nite. As a consequence, we will speak of continuous mappings of our
nite
groups, although this is strictly speaking super"uous: all mappings of a discrete topological
space are continuous. However, using the terminology as we do, it will
t perfectly in the
general framework of topological and Lie groups.

10.4 Let N 2 N , set !N = e2

i=N . We introduce
GN = h!N i = f!Nk
k = 1
: : :
N g:

Thus GN is the multiplicative cyclic subgroup of order N of the circle group T = fz 2 C :

jzj = 1g  C . The relative topology on GN , inherated from the Euclidean topology on C
is discrete.
Next C (G) denotes the vector space of the continuous complex valued functions on a
topological group G. Of course C (GN ) may be identi
ed with C N .
A character on a group G is a continuous homomorphism from G to T . The set of
characters on G form a group, denoted by G^ with multiplication
^ g 2 G:
e1 e2 (g) = e1 (g)e2(g)
ei 2 G

Ex. 10.5 Show that G^ is indeed a group. What is its unit element? Can you describe
the inversion?

APPLICATIONS OF FOURIER INTEGRALS

10.6 Lemma

57

G^ N 
= GN :

Proof De
ne : GN ! G^ N by

( (!Nj ))(!Nk ) = !Nkj :

One can check easily that is an injective homomorphism. It is also surjective: If e 2 G^ N ,
then e(!Nk ) = (e(!N ))k 2 T for all integers k. Since (e(!N ))N = e(id) = 1, we conclude
that e(!N ) = !Nj for some j 2 1
::
N . It then follows that e = (!Nj ).

10.7 As C (GN ) = C N , it can be given the inner product of C N . We want the function
1 to have norm 1, therefore we normalize:
X
F (g)G(g):
hF
Gi = N1
g2GN

We set ej = (!Nj ) 2 G^ N and sometimes we will identify ej and !Nj . We note that
!&N = !N;1 .

10.8 Proposition G^ N is an orthonormal basis of C (GN ).

Proof Since G^ N has N elements, which equals the dimension of C (GN ), it suces to
show that G^ N is an orthonormal system:

N
N
X
X
!Nlk !N;jk = N1 !N(l;j)k
k=1
( 1k=1
if l = j 
= !l;j 1;!N(l;l;jj)N = 0 if l 6= j .
N 1;!

hel
ej i = N1

The proposition implies the following representation formula for f 2 C (GN ):

N
X
X^
X
l
l
l
f (!N ) = hf
eie(!N ) = f (e)e(!N ) = f^(!Nj )!Nlj :
j =1
e2G^
e2G^

(10:5)

Here f^ is the nite Fourier transform of f . It is as expected, the function on G^ N de
ned

by
N
X
1
^
f (ej ) = (FN f )(ej ) = N f (!Nk )!N;jk :
(10:6)
k=1

10.9 For numerical purposes we are interested in the number of computations that are

necessary to compute f^(ej ). Formula ((10.6)) gives 2N computations to compute f^(ej ),

assuming knowledge of f and GN . Thus we need atmost cN 2 operations to compute FN f .
In fact we can do much better.

58

CHAPTER 10

10.3 Fast Fourier Transform

Theorem Suppose N = nm, n
m 2 N and let f 2 C (GN ). Let Fj 2 C (Gm ) be

de
ned by

Fj (!mk ) = f (!Nnk+j )
j = 1
: : :
n
k = 1
: : :
m

with !m = !Nn . Then we have
n
X
1
l
^
f (!N ) = n Fm(!N;lj Fj )(!ml mod m ):
j =1

Proof In the next computation we set p = nk + j , hence !Np = !mk !Nj . If p runs from 1
to N , then j runs from 1 to n and k from 1 to m.
N
n 1X
m
X
X
1
1
p
;
pl
l
^
f (!N ) = N f (!N )!N = n m Fj (!mk )(!m;k !N;j )l
p=1
j =1 k=1

!
n 1 X
m
X
1
!N;jl Fj (!mk )!m;kl =
=n m
j =1
k=1
n
X
= n1 (Fm !N;jl Fj )(!ml mod m):
j =1

10.10 Thus, to compute f^(!Nl ) we need to compute FmFj . Suppose it takes A(m)
computations to compute the Fourier transform in Gm (given F 2 C (Gm )). To compute
!N;jl Fj takes m computations, to compute Fm(!N;jl Fj ) takes A(m) + m computations.
We need all n of these, yielding n(A(m) + m) computations, the result of which we store.
Finally, for each f^(!Nl ) we need n additions of our stored results. This gives the total of
computations
A(N ) = Nn + n(A(m) + m) = N (n + 1) + A(m)n:
10.11 Theorem If N = n1    np , np = 1, then A(N )
N (1 + Pi (ni + 1)).
Proof Using the previous calculation we compute
A(n1    np )
N (n1 + 1) + A(n2    np)n1
N (n1 + 1) + N (n2 + 1) + A(n3    np )n2n1
X

: : :
N (ni + 1) + N:

10.12 Corollary If N = qp we need at most O(N log N ) computations to compute

Fourier transform in GN . (Here q is
xed and p ! 1.)
Proof
p
X
A(n)
N (2 (q + 1) + N ) = N (2p(q + 1) + 1):
j =1

Note that p = log N= log q, so that

N (q + 1) + N = O(N log N ):
A(N )
2N log
log q

APPLICATIONS OF FOURIER INTEGRALS

59

10.13 Next we indicate the relation of this
nite Fourier transform with ordinary Fourier
transform on the line. Suppose f has compact support on ;N
N ] and the values of f
at the points j=K are known, j 2 J = f;KN
;KN + 1
: : :
KN ; 1g. Then we can
approximate f^(x) as follows:
ZN
Z NK s
X
;ixj
1
ixt
f^(x) =
f (t)e dt = K
f ( K )e;i Kx sds  K1 f ( Kj )e K :
;N
;NK
j 2J

We substitute x = Np and
nd

1 X f ( j )e;i 22NKpj = 2N X 2 Jf ( j )!;pj = 2N (F g)(!p ):
f^( p
)

M
M
N K j2J K
2KN j
K M
j ) = f (j=K ), (j 2 J ).
Here we have written M = 2KN and de
ned g 2 C (GM ) by g(!M
Thus we have found a way to compute the (approximate) Fourier transform of a
function of which the value is known only in
nitely many points. Using the \fast way"
of computing it via Fourier transform on
nite groups as we described here is called Fast
Fourier Transform. The paper of Cooley and Tuckey on this topic is in fact in the top ten
of most cited mathematical papers.

Ex. 10.14 Convolution on GN For f
g 2 C (GN ) the convolution f  g is de
ned as

follows:

N
X
f  g(!Nj ) = N1 f (!Nk )g(!Nj;k ):

k=1
Show that FN (f  g)(!Nl ) = FN f (!Nl )FN g(!Nl ).

Also show that f  g = g  f and f  (g  h) = (f  g)  h.

Ex. 10.15 Fast multiplication Let F
G be polynomials of degree, p
q respectively.

View F and G as elements of C (GN ) with N = p + q + 1, by declaring F (!Nk ) equal to the

coecient of zk of F  the same goes for G.
Give a rough estimate for the number of computations needed to compute F  G
directly, and next, by observing that F  G as element of C (GN ) is the convolution of F
and G and using fast Fourier transform.
Apply this and describe a fast way for computing the product of two very large (10000digit) numbers.