Sie sind auf Seite 1von 71

Solutions to some exercises and problems

Teck-Cheong Lim
Department of Mathematical Sciences
George Mason University
4400, University Drive
Fairfax, VA 22030
U.S.A.
e-mail address: tlim@gmu.edu
Abstract
Solutions to some exercises and problems from Stein and Shakarchis
Fourier Analysis. The book by Y. Ketznelson, An introduction of Har-
monic Analysis (2nd corrected edition) is referred to frequently.
Chapter 1: The Genesis of Fourier Analysis
Chapter 2: Basic Properties of Fourier Series
Chapter 3: Convergence of Fourier Series
Chapter 4: Some applications of Fourier Series
Chapter 5: The Fourier transform on R
Chapter 6: The Fourier transform on R
d
Chapter 7: Finite Fourier Analysis
Chapter 8: Dirichlets Theorem
Chapter 1 The Genesis of Fourier Analysis
1. (Exercise 8) Suppose F is a function on (a, b) with two continuous deriva-
tives. Show that whenever x and x +h belong to (a, b), one may write
F(x +h) = F(x) +hF

(x) +
h
2
2
F

(x) +h
2
(h),
where (h) 0 as h 0.
Deduce that
F(x +h) +F(x h) 2F(x)
h
2
F

(x) as h 0.
Proof.
Firstly one has
F(x +h) F(x) =
_
x+h
x
F

(y) dy.
1
Then F

(y) = F

(x)+(yx)F

(x)+(yx)(yx), where is continuous


and (h) 0 as h 0. Therefore
F(x +h) F(x) = F

(x)h +F

(x)
h
2
2
+
_
h
0
u(u) du.
By mean-value theorem
_
h
0
u(u) du = ()
_
h
0
udu = h
2
()
2
= h
2
(h)
where is between 0 and h, and (h) 0 as h 0 by the continuity of
.
From above, one easily gets
F(x +h) +F(x h) 2F(x)
h
2
F

(x) = (h) +(h) 0 as h 0.


Remark. Suppose F is periodic. The function (x, h) =
F(x+h)+F(xh)2F(x)
h
2

F

(x) is continuous for all x, h. And there exists M such that


[F(x +h) +F(x h) 2F(x)[ Mh
2
for all x, h (since F, F

, are all periodic in x).


2. (Exercise 10) Show that the expression of the Laplacian
=

2
x
2
+

2
y
2
is given in polar coordinates by the formula
=

2
r
2
+
1
r

r
+
1
r
2

2
.
Also prove that

u
x

2
+

u
y

2
=

u
r

2
+
1
r
2

2
.
Solution.
From x = r cos , y = r sin , we get
_
x
r
x

y
r
y

_
=
_
cos r sin
sin r cos
_
,
hence
_
r
x
r
y

y
_
=
_
cos r sin
sin r cos
_
1
=
_
cos sin

1
r
sin
1
r
cos
_
.
2
By Chain Rule,
u
x
=
u
r
r
x
+
u

x
=
u
r
cos
1
r
u

sin .
By Chain Rule and Product Rule,

x
_
u
r
cos
_
=

r
_
u
r
cos
_
r
x
+

_
u
r
cos
_

x
=

2
u
r
2
cos
2
+
1
r
u
r
sin
2

1
r

2
u
r
cos sin
Similarly,

x
_

1
r
u

sin
_
=
1
r
2

2
u

2
sin
2
+
2
r
2
+
2
r
2
u

sin cos
1
r

2
u
r
cos sin .
Since

2
u
r
=

2
u
r
, we get

2
u
x
2
=

2
u
r
2
cos
2
+
1
r
u
r
sin
2
+
1
r
2

2
u

2
sin
2
+
2
r
2
u

sin cos
2
r

2
u
r
cos sin .
Similarly,
u
y
=
u
r
sin +
1
r
u

cos ,

2
u
y
2
=

2
u
r
2
sin
2
+
1
r
u
r
cos
2
+
1
r
2

2
u

2
cos
2

2
r
2
u

sin cos +
2
r

2
u
r
cos sin .
Consequently,

2
u
x
2
+

2
u
y
2
=

2
u
r
2
+
1
r
u
r
+
1
r
2

2
u

2
.
Assume the general case that u is complex-valued. Then

u
x

2
=

u
r

2
cos
2
+
1
r
2

2
sin
2

2
r
1
u
r
u

cos sin
and

u
y

2
=

u
r

2
sin
2
+
1
r
2

2
cos
2
+
2
r
1
u
r
u

cos sin ,
hence

u
x

2
+

u
y

2
=

u
r

2
+
1
r
2

2
.
3
3. (Added in) Let f(x) be an odd 2-periodic function. For n Z dene
c
n
=
1
2
_

f(t)e
int
dt
Prove that
cne
inx
+c
n
e
inx
= A
n
sin nx
where
A
n
=
2

_

0
f(t) sin nt dt.
Proof.
cne
inx
+c
n
e
inx
=
1
2
_

f(t)e
int
e
inx
+f(t)e
int
e
inx
dt
=
1
2
_

f(t)(e
in(xt)
+e
in(xt)
) dt
=
1
2
_

f(t)2 cos n(x t) dt


=
1

f(t)(cos nxcos nt + sin nxsin nt) dt


=
1

f(t) sin nxsin nt dt (since f(t) cos nt is odd)


= sin nx
2

_

0
f(t) sin nt dt (since f(t) sin nt is even)
= A
n
sin nx
4. (Exercise 11) Show that if n Z the only solutions of the dierential
equation
r
2
F

(r) +rF

(r) n
2
F(r) = 0,
which are twice dierentiable when r > 0, are given by linear combina-
tions of r
n
and r
n
when n ,= 0, and 1 and log r when n = 0.
Solution.
If F(r) solves the equation, write g(r) = F(r)/r
n
. Then g is twice dier-
entiable, F(r) = r
n
g(r) and
F

(r) = nr
n1
g(r) +r
n
g

(r),
F

(r) = n(n 1)r


n2
g(r) +nr
n1
g

(r) +nr
n1
g

(r) +r
n
g

(r).
Then
r
2
F

(r)+rF

(r)n
2
F(r) = r
n+1
(rg

(r)+(2n+1)g

(r)) = r
n+1
(2ng

(r)+(rg

(r))

) = 0,
4
hence
2ng(r) +rg

(r) = c.
It follows that g(r) is a linear combination of r
2n
and 1 if n ,= 0, and
log r and 1 if n = 0. The result follows.
5. (Problem 1) Consider the Dirichlet problem illustrated in Figure 11. More
precisely, we look for a solution of the steady-state heat equation u = 0
in the rectangle R = (x, y) : 0 x , 0 y 1 that vanishes on the
vertical sides of R, and so that
u(x, 0) = f
0
(x) and u(x, 1) = f
1
(x),
where f
0
and f
1
are initial data which x the temperature distribution on
the horizontal sides of the rectangle.
Use separation of variables to show that if f
0
and f
1
have Fourier expan-
sions
f
0
(x) =

k=1
A
k
sin kx and f
1
(x) =

k=1
B
k
sin kx,
then
u(x, y) =

k=1
_
sinh k(1 y)
sinh k
A
k
+
sinh ky
sinh k
B
k
_
sin kx.
Compare this result with the solution of the Dirichlet problem in the strip
obtained in Problem 3, Chapter 5.
Solution.
Consideration of basic solution of the form A(x)B(y) yields
A(x) = sin kx, B(y) = c
k
sinh ky +d
k
cosh ky.
Setting y = 0, y = 1 and comparing coecients give
d
k
= A
k
, c
k
sinh k +d
k
cosh k = B
k
.
Solving for d
k
, c
k
and using the identity sinh(a b) = sinh a cosh b
sinh b cosh a, we nd
c
k
sinh ky +d
k
cosh ky =
sinh k(1 y)
sinh k
A
k
+
sinh ky
sinh k
B
k
.
Chapter 2
1. (Exercise 1) Suppose f is 2-periodic and integrable on any nite interval.
Prove that if a, b R, then
_
b
a
f(x) dx =
_
b+2
a+2
f(x) dx =
_
b2
a2
f(x) dx.
5
Also prove that
_

f(x +a) dx =
_

f(x) dx =
_
+a
+a
f(x) dx.
Proof.
For the rst two identities, substitute u = x 2 and u = x + 2 respec-
tively, and note that f(x 2) = f(x). For the second equalities, rstly
we have
_

f(x+a) dx =
_
+a
+a
f(x) dx by substitution u = x+a. Then
by applying the rst set of equalities, we get
_

+a
f(x) dx =
_

+a
f(x) dx,
so
_
+a
+a
f(x) dx =
_

+a
f(x) dx+
_

f(x) dx+
_
+a

f(x) dx =
_

f(x) dx.
2. (Exercise 2) In this exercise we show how the symmetries of a function
imply certain properties of its Fourier coecients. Let f be a 2-periodic
Riemann integrable function on R.
(a) Show that the Fourier series of the function f can be written as
f()

f(0) +

n1
[

f(n) +

f(n)] cos n +i[

f(n)

f(n)] sin n.
(b) Prove that if f is even, then

f(n) =

f(n), and we get a cosine series.
(c) Prove that if f is odd, then

f(n) =

f(n), and we get a sine series.


(d) Suppose that f( + ) = f() for all R. Show that

f(n) = 0 for
all odd n.
(e) Show that f is real-valued if and only if

f(n) =

f(n) for all n.
Proof.

f(n)e
int
+

f(n)e
int
= (

f(n) +

f(n))
1
2
(e
int
+e
int
) +i(

f(n)

f(n))
1
2i
(e
int
e
int
)
= A
n
cos nt +B
n
sin nt
where A
n
=

f(n)+

f(n) =
1
2
_
2
0
f(s)(e
ins
+e
ins
) ds =
1

_
2
0
f(s) cos ns ds
and B
n
= i(

f(n)

f(n)) = i
1
2
_
2
0
f(s)(e
ins
e
ins
) ds =
1

_
2
0
f(s) sin ns ds.
If f is real, then A
n
, B
n
are real, which is true i

f(n) =

f(n) for all n.
If f is odd, then f(s) cos ns is odd, so A
n
= 0. If f is even, then f(s) sin ns
is odd, so B
n
= 0.
If f( +) = f() for all R, then
_
2
0
f(t)e
int
dt =
_

0
f(t)e
int
dt +
_
2

f(t)e
int
dt
6
and
_
2

f(t)e
int
dt =
_

0
f(u +)e
inu
e
in
du =
_

0
f(t)e
int
dt
for all odd n.
3. (Exercise 4) Consider the 2-periodic odd function dened on [0, ] by
f() = ( ).
(a) Draw the graph of f.
(b) Computer the Fourier coecients of f, and show that
f() =
8

k odd 1
sin k
k
3
Solution.
Checked with Maple and found that it is correct.
4. (Exercise 6) Let f be the function dened on [, ] by f() = [[.
(a) Draw the graph of f.
(b) Calculate the Fourier coecients of f, and show that ... O(1/n
2
).
(c) What is the Fourier series of f in terms of sines and cosines?
(d) Taking = 0, prove that

n odd 1
1
n
2
=

2
8
and

n=1
1
n
2
=

2
6
.
Solution.
Since it is an even function, by Exercise 2,

f(n) =

f(n). We nd

f(n) +

f(n) = 0, if n is even, and


2
n
2
if n is odd. So the series is

2

4

n odd 1
cos n
n
2
.
Assuming that the series converges at = 0, we nd the rst part of (d).
It is easy to see that

n even 1
1
n
2
=
1
4

n=1
1
n
2
. Combining this with
what we got for the odd sum, we get the second part of (d).
5. (Exercise 7) Suppose a
n

N
n=1
and b
n

N
n=1
are two nite sequences of
complex numbers. Let B
k
=

k
n=1
b
n
denote the partial sums of the
series

b
n
with the convention B
0
= 0.
(a) Prove the summation by parts formula
N

n=M
a
n
b
n
= a
N
B
N
a
M
B
M1

N1

n=M
(a
n+1
a
n
)B
n
.
7
(b) Deduce from this formula Dirichlets test for convergence of a series :
if the partial sums of the series

b
n
are bounded, and a
n
is a sequence
of real numbers that decreases monotonically to 0, then

a
n
b
n
converges.
Proof. (a)
N

n=M
a
n
b
n
=
N

n=M
a
n
(B
n
B
n1
)
=
N

n=M
a
n
B
n

n=M
a
n
B
n1
=
N

n=M
a
n
B
n

N1

n=M1
a
n+1
B
n
= a
n
B
N
+
N1

n=M
a
n
B
n
a
M
B
M1

N1

n=M
a
n+1
B
n
= a
N
B
N
a
M
B
M1

N1

n=M
(a
n+1
a
n
)B
n
(b) Suppose [B
n
[ B for all n. Then for N > M
[
N

n=M
a
n
b
n
[ B(a
N
+a
M
+
N1

n=M
(a
n
a
n+1
)) = 2Ba
M
proving that the series is Cauchy and hence converges.
6. (Exercise 8) Verify that
1
2i

n=0
e
inx
n
is the Fourier series of the 2-
periodic sawtooth function illustrated in Figure 6, dened by f(0) = 0,
and
f(x) =
_
/2 x/2 if < x < 0
/2 x/2 if 0 < x < .
Note that this function is not continuous. Show that nevertheless, the
series converges for every x (by which we mean, as usual, that the sym-
metric partial sums of the series converge). In particular, the value of the
series at the origin, namely 0, is the average of the values of f(x) as x
approaches the origin from the left and the right.
Solution.
Checked with Maple and found it correct. b
n
= sin nx,
B
n
=
n

k=1
b
k
=
sin(nx/2) sin((n + 1)x/2)
sin(x/2)
8
(to prove write sin nx as
1
2i
(e
inx
e
inx
))) and [B
n
[ csc([x[/2) for
[x[ < , x ,= 0. 1/n is decreasing to 0. So series converges for x ,= 0. At
x = 0, all the symmetric sums are 0, so series converges to 0. It must
converge to the function since its Cesaro means converge to the function.
7. (Exercise 9) Let f(x) =
[a,b]
(x) be the characteristic function of the
interval [a, b] [, ], that is,

[a,b]
(x) =
_
1 if x [a, b]
0 otherwise.
(a) Show that the Fourier series of f is given by
f(x)
b a
2
+

n=0
e
ina
e
inb
2in
e
inx
.
The sum extends over all positive and negative integers excluding 0.
(b) Show that if a ,= or b ,= and a ,= b, then the Fourier series does
not converge absolutely for any x.
(c) However, prove that Fourier series converges at every point x. What
happens if a = and b = ?
Solution.
(a) Straightforward.
(b) [e
ina
e
inb
[
2
= 1 + 1 2 cos n(b a) = 4 sin
2
(n(b a)/2), thus
[e
ina
e
inb
[ = 2[ sin n
0
[, where
0 <
0
=
b a
2
<

2
.
The series

n=1
[ sin n
0
[
n
diverges because .......
(c) a
n
= 1/n, b
n
= sin(na) sin(nb) which has bounded partial sums by
the solution of Exercise 8. So series converges by Exercise 7. If a =
and b = , then the Fourier series is 1.
8. (Exercise 10) Suppose f is a periodic function of period 2 which belongs
to the class C
k
. Show that

f(n) = O(1/[n[
k
) as [n[ .
Solution.
Integrate by parts k times as in Corollary 2.4.
9. (Exercise 11) Suppose that f
k

k=1
is a sequence of Riemann integrable
functions on the interval [0, 1] such that
_
1
0
[f
k
(x) f(x)[ dx 0 as k .
9
Show that

f
k
(n)

f(n) uniformly in n as k .
Proof.
Use
[ g(n)[
_
1
0
[g(x)[ dx
for all n.
10. (Exercise 12) Prove that if a series of complex numbers

c
n
converges to
s, then

c
n
is Cesaro summable to s.
Proof.
Let the sequence of partial sums be s
n
, n = 1, 2, . First assume that
s = 0. Let > 0. Choose N
1
such that [s
n
[ <

2
for all n N
1
. Choose
N > N
1
such that

N1
k=1
[s
k
[
n
<

2
for all n N. Then for all n N,
[
s
1
+ +s
n
n
[

[s
1
[ + +[s
n
[
n
=
[s
1
+ +[s
N1
[
n
+
[s
N1+1
+ +[s
n
[
n
<

2
+
n/2
n
=
This proves that

c
n
is Cesaro summable to 0. Notice that we prove
above that if a
n
is any sequence converging to 0, then the sequence
n
=
a1++an
n
converges to 0.
Now suppose that s ,= 0. Then sequence t
n
= s
n
s converges to 0. So
by the proof above (t
1
+ +t
n
)/n = (s
1
+ +s
n
)/n s converges to
0, i.e (s
1
+ +s
n
)/n converges to s.
11. (Exercise 13) The purpose of this exercise is to prove that Abel summa-
bility is more general than the standard or Cesaro methods of summation.
(a) Show that if the series

n=1
c
n
of complex numbers converges to a
nite limit s, then the series is Abel summable to s.
(b) However, show that there exist series which are Abel summable, but
that do not converge.
(c) Argue similarly to prove that if a series

n=1
c
n
is Cesaro summable
to , then it is Abel summable to .
(d) Give an example of a series that is Abel summable but not Cesaro
summable.
The results above can be summarized by the following implications about
series:
convergent Cesaro summable Able summable,
10
and the fact that none of the arrows can be reversed.
Solution:
In what follows, 0 r < 1; s
n
denotes the sequence of partial sums of

n=1
c
n
.
(a). First assume that s = 0. For 0 r < 1 consider t
n
=

n
k=1
s
k
r
k
.
We have rt
n
=

n
k=1
s
k
r
k+1
=

n+1
k=2
s
k1
r
k
. Since s
1
= c
1
and s
k

s
k1
= c
k
, we get (1 r)t
n
=

n
k=1
c
k
r
k
s
n
r
n+1
, hence

n
k=1
c
k
r
k
=
(1 r)

n
k=1
s
k
r
k
+ s
n
r
n+1
. Since c
k
, s
k
are bounded sequences (in fact,
they converge to 0) and 0 r < 1, both series

k=1
c
k
r
k
and

k=1
s
k
r
k
are absolutely convergent, and

k=1
c
k
r
k
= (1r)

k=1
s
k
r
k
. Let > 0.
Choose K such that [s
k
[ < for all k K. Then
[

k=1
c
k
r
k
[ (1 r)
K1

k=1
Mr
k
+ (1 r)
r
K
1 r
(1 r)
K1

k=1
Mr
k
+
It follows that
limsup
r1
[

k=1
c
k
r
k
[ .
Since > 0 is arbitrary, we have limsup
r1
[

k=1
c
k
r
k
[ 0 which is
equivalent to lim
r1

k=1
c
k
r
k
= 0.
Next assume that s ,= 0. Dene d
n
= c
n
s/2
n
. Then

n=1
d
n
= 0,
and

n=1
d
n
r
n
=

n=1
c
n
r
n
sr/(2 r). Thus by our proof above,
lim
r1

n=1
c
n
r
n
s = 0, i.e. lim
r1

n=1
c
n
r
n
= s.
(b). The series

n=1
(1)
n
does not converge, but lim
r1

n=1
(1)
n
r
n
=
lim
r1
r
1+r
= 1/2, i.e.

n=1
(1)
n
is Abel summable to 1/2.
(c). Write

n
=
s
1
+ +s
n
n
and
n
= n
n
. Since
n

n1
= s
n
, by the same argument as in (a), we
have
(1 r)

k=1

k
r
k
=

k=1
s
k
r
k
,
hence
(1 r)
2

k=1

k
kr
k
=

k=1
c
k
r
k
from the proof of (a). Assume that = 0. Let > 0. Choose N such that
[
k
[ < for all n N, and B be a bound for [
k
[, k = 1, 2, . Then
[

k=1
c
k
r
k
[ (1r)
2
N1

k=1
Bkr
k
+(1r)
2

r
N
((1 r)N +r)
(1 r)
2
(1r)
2
N1

k=1
Bkr
k
+r
N
((1r)N+r)
where we have use the fact that

k=N
kr
k
=
r
N
((1r)N+r)
(1r)
2
. As in the
proof of (a), this implies that lim
r1

k=1
c
k
r
k
= 0. For the case ,= 0,
11
consider d
n
= c
n
/2
n
. Since

n=1
c
n
is Cesaro summable to and

n=1
/2
n
= (and hence Cesaro summable to ), we have

n=1
d
n
Cesaro summable to 0. Thus by our proof above and the same argument
as in the proof of (a), the result follows.
(d). Consider the series

n=1
(1)
n
n. It is Abel summable to 1/4 since

n=1
(1)
n
nr
n
=
r
(1 +r)
2
.
Note that

n 1
n

n1
=
a
n
n
.
Thus for a Cesaro summable series

n=1
a
n
, lim
n
an
n
must be 0. This
proves that

n=1
(1)
n
n is not Cesaro summable.
12. (Exercise 14) This exercise deals with a theorem of Tauber which says that
under an additional condition on the coecients c
n
, the above arrows can
be reversed.
(a) If

c
n
is Cesaro summable to and c
n
= o(1/n) (that is, nc
n
0),
then

c
n
converges to .
(b) The above statement holds if we replace Cesaro summable by Abel
summable.
Proof.
(a). Denote (n 1)c
n
by t
n
. Since t
n
=
n1
n
nc
n
and nc
n
0 we have
t
n
0. Now
s
n

n
= s
n

s
1
+ +s
n
n
=
(s
n
s
1
) + + (s
n
s
n
)
n
=
t
2
+ +t
n
n
and it is immediate that s
n

n
0.
(b). Let r = 1
1
N
. We have
[
N

n=1
c
n

n=1
c
n
r
n
[
N

n=1
[c
n
[
_
1 (1
1
N
)
n
_

n=1
[c
n
[
n
N
where we have used (1
1
N
)
n
1
n
N
for 1 n N, which can be proved
12
easily by induction on n. Also if [c
n
n[ < for all n N, then
[

n=1
c
n
r
n

n=1
c
n
r
n
[

n=N+1
[c
n
[(1
1
N
)
n

n=N+1
n[c
n
[
N
(1
1
N
)
n

n=N+1

N
(1
1
N
)
n
= (1
1
N
)
N+1


e
as N .
13. (Exercise 15) Prove that the Fejer kernel is given by
F
N
(x) =
1
N
sin
2
(Nx/2)
sin
2
(x/2)
.
Proof.
Recall that NF
N
(x) = D
0
(x)+ +D
N1
(x) where D
n
(x) is the Dirichlet
kernel. Write = e
ix
. Then
D
n
(x)
=
n
+ +
1
+ 1 + + +
n
= (
n
+ +
1
) + (1 + + +
n
)
=
1

n
1

1
1
+
1
n+1
1
=

n
1
1
+
1
n+1
1
=

n

n+1
1
13
So
NF
N
(x)
=
N1

n=0

n+1
1
=
1
1
_
N1

n=0

N1

n=0

n+1
_
=
1
1
_

N
1

1
1

1
N
1
_
=
1
1
_

N+1

1

1
N
1
_
=

N
2 +
N
(1 )
2
=
1
(
1/2
)
2
(
N/2

N/2
)
2
(1 )
2
=
(
N/2

N/2
)
2
(
1/2

1/2
)
2
=
4 sin
2
(Nx/2)
4 sin
2
(x/2)
=
sin
2
(Nx/2)
sin
2
(x/2)
.
Therefore
F
N
(x) =
1
N
sin
2
(Nx/2)
sin
2
(x/2)
.
14. (Exercise 16) The Weierstrass approximation theorem states: Let f be a
continuous function on the closed and bounded interval [a, b] R. Then,
for any > 0, there exists a polynomial P such that
sup
x[a,b]
[f(x) P(x)[ < .
Proof.
Let > 0. We may extend f to a continuous (ca)-periodic function where
b c. By Corollary 5.4 of Fejers theorem, there exists a trigonometric
polynomial Q =

N
n=M
a
n
e
inx
such that [Q(x)f(x)[ < /2 for all x. For
each n, M n N, there exists a polynomial p
n
(x) such that [a
n
e
inx

p
n
(x)[ < /2N for all x [a, c]. Then P = p
M
+ +p
N
is a polynomial
in x that satises the requirement.
15. (Exercise 17) In Section 5.4 we proved that the Abel means of f converge
to f at all points of continuity, that is,
lim
r1
A
r
(f)() = lim
r1
(P
r
f)() = f(), with 0 < r < 1,
14
whenever f is continuous at . In this exercise, we will study the behavior
of A
r
(f)() at certain points of discontinuity.
An integrable function is said to have a jump discontinuity at if the
two limits
lim
h0,h>0
f( +h) = f(
+
) and lim
h0,h<0
f( +h) = f(

)
exist.
(a) Prove that if f has a jump discontinuity at , then
lim
r1
A
r
(f)() =
f(
+
) +f(

)
2
, with 0 r < 1.
(b) Using a similar argument, show that if f has a jump discontinuity at
, the Fourier series of f at is Cesaro summmable to
f(
+
)+f(

)
2
.
Proof.
Since P
r
() = P
r
(), and
1
2
_

P
r
() d = 1, we have
1
2
_
0

P
r
() d =
1
2
_

0
P
r
() d = 1/2. Suppose f has a jump discontinuity at . Let > 0
be given. Choose > 0 so that 0 < h < implies [f( h) f()[ <
and [f( + h) f(+)[ < . Let M be such that [f(y)[ M for all y.
Then

(f P
r
)()
f(
+
) +f(

)
2

_
1
2
_

P
r
(y)f( y) dy
_

f(
+
) +f(

)
2

1
2
_
0

P
r
(y)[f( y) f(
+
)[ dy +
1
2
_

0
P
r
(y)[f( y) f(

)[ dy

1
2
_
<y<0
P
r
(y)[f( y) f(
+
)[ dy +
1
2
_
0<y<
P
r
(y)[f( y) f(

)[ dy
+
1
2
2M
_
|y|
P
r
(y) dy


2
+

2
+
M

_
|y|
P
r
(y) dy
Therefore, recalling that lim
r1
_
|y|
P
r
(y) dy = 0,
limsup
r1

(f P
r
)()
f(
+
) +f(

)
2


Since > 0 is arbitrary, we have
lim
r1
(f P
r
)() =
f(
+
) +f(

)
2
.
(b) Since the Fejer kernel F
n
() is even and positive, we also have
1
2
_
0

F
n
() d =
1
2
_

0
F
n
() d = 1/2, for all n. Repeat the above argument.
15
Remark One can replace
f(
+
)+f(

)
2
by lim
h0
f(+h)+f(h)
2
and obtain
a more general statement of Exercise 17. (See Katznelsons book.)
16. (Exercise 18) If P
r
() denotes the Poisson kernel, show that the funciton
u(r, ) =
P
r

,
dened for 0 r < 1 and R, satises:
(i) u = 0 in the disc.
(ii) lim
r1
u(r, ) = 0 for each .
However, u is not identically zero.
Solution.
We have
u(r, ) =

n=
r
|m|
ine
in
.
Each summand is Laplacian (see chapter 1, separation of variable), and the
series and its derivatives converge uniformly on r < for any 0 < < 1,
hence the innite sum is also Laplacian.
u(r, ) =
2r(r
2
1) sin
(1 2r cos +r
2
)
2
If ,= 0, then 1 2 cos +1 ,= 0, so the limit is 0 as r 1. If = 0, then
u(r, 0) = 0 and the limit is trivially 0. Note that
u(x, y) =
2y(x
2
+y
2
1)
((1 x)
2
+y
2
)
2
and u(1 , ) as 0+. This implies that u does not converge
to 0 uniformly as r 1.
17. (Exercise 19) Solve Laplaces equation u = 0 in the semi innite strip
S = (x, y) : 0 < x < 1, 0 < y,
subject to the following boundary conditions
_
_
_
u(0, y) = 0 when 0 y,
u(1, y) = 0 when 0 y,
u(x, 0) = f(x) when 0 x 1
where f is a given function, with of course f(0) = f(1) = 0. Write
f(x) =

n=1
a
n
sin(nx)
and expand the general solution in terms of the special solutions given by
u
n
(x, y) = e
ny
sin(nx).
16
Express u as an integral involving f, analogous to the Poisson integral
formula (6).
Solution.
By considering the odd extension of f and following the derivation of
Poissons kernel with e
y
and e
it
replacing r and e
it
, respectively, we
obtain
u(x, y) =
1
2
_
1
1
f(t)Q
y
(x t) dt
where
Q
y
(t) =
1 e
2y
1 2e
y
cos t +e
2y
.
or, using the fact that f is odd, we have the alternate form
u(x, y) =
1
2
_
1
0
f(t)Q(x, t) dt
where
Q(x, t) =
1 e
2y
1 2e
y
cos (x t) +e
2y

1 e
2y
1 2e
y
cos (x +t) +e
2y
.
18. (Exercise 20) Consider the Dirichlet problem in the annulus dened by
(r, ) : < r < 1, where 0 < < 1 is the inner radius. The problem is
to solve

2
r
2
+
1
r

r
+
1
r
2

2
= 0
subject to the boundary conditions
_
u(1, ) = f(),
u(, ) = g(),
where f and g are given continuous functions.
Arguing as we have previously for the Dirichlet problem in the disc, we
can hope to write
u(r, ) =

c
n
(r)e
in
with c
n
(r) = A
n
r
n
+B
n
r
n
, n ,= 0. Set
f()

a
n
e
in
and g()

b
n
e
in
.
We want c
n
(1) = a
n
and c
n
() = b
n
. This leads to the solution
u(r, ) =

n=0
_
1

n
_
[((/r)
n
(r/)
n
)a
n
+ (r
n
r
n
)b
n
]e
in
+a
0
+ (b
0
a
0
)
log r
log
.
17
Show that as a result we have
u(r, ) (P
r
f)() 0 as r 1 uniformly in ,
and
u(r, ) (P
/r
g)() 0 as r uniformly in .
Solution.
Note that
r
n
r
n

n
=
r
2n
1

2n
1
(/r)
n
0 and
(/r)
n
(r/)
n

n
=
(/r)
2n
1

2n
1
r
n

0 as n .
The series converges because for instance
r
2n
1

2n
1
(/r)
n

1 +r
2
1
2
(/r)
n
.
And the last result follows from, for instance,
_
(/r)
2n
1

2n
1
1
_
r
n
=
_
(/r)
2n

2n

2n
1
_
r
n
2[(/r)
2n

2n
]r
n
and

n=1
[(/r)
2n

2n
]r
n
=

2
/r
1
2
/r


2
r
1
2
r
which approaches 0 as r 1. Details are to be given.
19. (Problem 2) Let D
N
denote the Dirichlet kernel
D
N
() =
N

n=
N
e
ik
=
sin(N + 1/2)
sin(/2)
,
and dene
L
N
=
1
2
_

[D
N
()[ d.
(a) Prove that
L
N
c log N
for some constant c > 0. A more careful estimate gives
L
N
=
4

2
log N +O(1).
(See Katznelson Chapter 2, Exercise 1.1 for section p.50.)
(b) Prove the following as a consequence: for each n 1, there exists a
continuous function f
n
such that [f
n
[ 1 and [S
n
(f
n
)(0)[ c

log n.
Solution.
18
(a) Note that
x
sin x
1 for x in the interval [/2, /2]. It follows that
for [, ],
[D
N
()[ 2
[ sin(N + 1/2)[

.
Then
_

[D
N
()[ d
4
_

0
[ sin(N + 1/2)[

d
= 4
_
(N+1/2)
0
[ sin [

d
4
_
N
0
[ sin [

d
= 4
N1

k=0
_
(k+1)
k
[ sin [

d
4
N1

k=0
1
(k + 1)
_
(k+1)
k
[ sin [ d
=
8

N1

k=0
1
k + 1
=
8

log(N + 1)

log N
Therefore L
N
c log N.
(b) The function g
n
which is equal to 1 when D
n
is positive and 1 when
D
n
is negative has the desired property but is not continuous. Approxi-
mate g
n
in the integral norm (in the sense of Lemma 3.2) by continuous
functions h
k
satisfying [h
k
[ 1.
20. (Problem 3) Littlewood provided a renement of Taubers theorem:
(a) If

c
n
is Abel summable to s and c
n
= O(1/n), then

c
n
converges
to s.
(b) As a consequence, if

c
n
is Ces`aro summable to s and c
n
= O(1/n),
then

c
n
converges to s.
These results may be applied to Fourier series. By Exercise 17, they imply
that if f is an integrable function that satises

f() = O(1/[[), then:
(i) If f is continuous at , then
S
N
(f)() f() as N .
(ii) If f has a jump discontinuity at , then
S
N
(f)()
f(
+
) +f(

)
2
as N .
19
(iii) If f is continuous on [, ], then S
N
(f) f uniformly.
For the simpler assertion (b), hence of proof of (i),(ii),and (iii), see Problem
5 in Chapter 4.
Solution.
Lemma 1 Let f be a C
2
real function on [0, 1). Suppose as x 1,
f(x) = o(1), f

(x) = O
_
1
(1x)
2
_
, then f

(x) = o
_
1
1x
_
.
Proof. Choose C > 0 such that [(1 x)
2
f

(x)[ C for all x. Let


> 0. Choose a , 0 < < min1/2, /(4C). Choose < 1 such that
[f(x)[ < (1/4) for all x > . We claim that [(1 x)f

(x)[ < for all


x > . Let x > . With x

= x +(1 x) we have
f(x

) = f(x) +(1 x)f

(x) +
1
2

2
(1 x)
2
f

()
for some x < < x

.
Since < 1/2, we have 1 x 2(1 ), so that [(1 x)
2
f

()[
4[(1 )
2
f

()[ 4C. Therefore


[(1 x)f

(x)[
= [
f(x

) f(x)


1
2
(1 x)
2
f

()[

[f(x

)[ +[f(x)[

+
1
2
(1 x)
2
[f

()[
<
1
2
+
1
2
=
Lemma 2 Let

n=0
a
n
x
n
be a real power series with a
n
0 for all n.
Suppose
lim
x1

(1 x)

n=0
a
n
x
n
= 1.
Then for any integrable function g(t) on [0, 1]
lim
x1

(1 x)

n=0
a
n
x
n
g(x
n
) =
_
1
0
g(t) dt.
Proof. First prove for functions of the type x
k
. Then for any polynomial;
then for any continuous function; then for any integral function.
Corollary 1 Let

n=0
a
n
x
n
be a real power series with a
n
0 for all n.
Suppose
lim
x1

(1 x)

n=0
a
n
x
n
= 1.
Then
lim
N

N
n=0
a
n
N
= 1.
20
Proof. Apply the above lemma to the function g(t) = 0 for t < e
1
, 1/t
otherwise, and let x = e
1/N
.
Finishing the solution of Problem 3
We shall write f(x) g(x) to mean
f(x)
g(x)
1 as x 1, and f(n) g(n)
to mean
f(n)
g(n)
1 as n .
(a) We may assume that s = 0 (see Exercise 13, Chapter 2), i.e. we assume
that f(x) =

n=0
a
n
x
n
0 as x 1. Then f

(x) = O(1/(1 x)
2
)
because
f

(x) =

n=2
n(n 1)a
n
x
n2
= O(

n=2
(n 1)x
n2
) = O(1/(1 x)
2
).
So by Lemma 1,
f

(x) = o
_
1
1 x)
_
.
Suppose [na
n
[ c. Then

n=1
(1
na
n
c
)x
n1
=
1
1 x

f

(x)
c

1
1 x
Since 1
nan
c
0, Corollary 1 implies that
n

k=1
(1
ka
k
c
) n
or what is the same
n

k=1
ka
k
= o(n).
21
Write w
n
=

n
k=1
ka
k
, w
0
= 0. So w
n
/n 0 as n . Then
f(x) a
0
=

n=1
a
n
x
n
=

n=1
w
n
w
n1
n
x
n
=

n=1
w
n
n
x
n

n=1
w
n1
n
x
n
=

n=1
w
n
n
x
n

n=1
w
n
n + 1
x
n+1
=

n=1
w
n
_
x
n
n

x
n+1
n + 1
_
=

n=1
w
n
_
x
n
x
n+1
n + 1
+
x
n
n(n + 1)
_
= (1 x)

n=1
w
n
x
n
n + 1
+

n=1
w
n
n(n + 1)
x
n
Since f(x) 0 and the rst term in the last sum approaches 0 as x 1,
we get
lim
x1

n=1
w
n
n(n + 1)
x
n
= a
0
Since
wn
n(n+1)
= o(1/n), by the regular Tauberian theorem

n=1
w
n
n(n + 1)
= a
0
.
Now
N

n=1
w
n
n(n + 1)
=
N

n=1
w
n
_
1
n

1
n + 1
_
=
N

n=1
w
n
w
n1
n

w
N
N + 1
=
N

n=1
a
n

w
N
N + 1
.
Letting N we get

n=1
a
n
= a
0
, i.e.

n=0
a
n
= 0.
22
Chapter 3
1. (Exercise 1) Show that the rst two examples of inner product spaces,
namely R
d
and C
d
, are complete.
2. (Exercise 2) Prove that the vector space
2
(Z) is complete.
3. (Exercise 3) Construct a sequence of integrable functions f
k
on [0, 2]
such that
lim
k
1
2
_
2
0
[f
k
(t)[
2
dt = 0
but lim
k
f
k
(t) fails to exist for any t.
4. (Exercise 4) (In (c), use the fact that f is continuous except possibly on
a set of measure 0.)
5. (Exercise 5) Let
f(t) =
_
0 for t = 0
log(1/t) for 0 < t 2,
and dene a sequence of functions in ! by
f
n
(t) =
_
0 for 0 t 1/n
f(t) for 1/n < t 2.
Prove that f
n

n=1
is a Cauchy sequence in !. However, f does not be-
long to !.
Solution.
_
(log t)
2
dt = t(log t)
2
2t log t + 2t
and lim
t0
t log t = 0, lim
t0
t log
2
t = 0.
6. (Exercise 6) Consider the sequence a
k

k=
dened by
a
k
=
_
1/k if k 1
0 if k 0.
Note that a
k
l
2
(Z), but that no Riemann integrable function has k
th
Fourier coecient equal to a
k
for all k.
Solution.
Let M = sup
t
[f(t)[. Recall that
A
r
(f)() =
1
2
_

P
r
( t)f(t) dt.
So
[A
r
(f)()[ M
1
2
_

P
r
( t) dt = M
23
Also
A
r
(f)() =

n=

f(n)e
in
r
|n|
.
So
A
r
(f)(0) =

n=

f(n)r
|n|
.
It follows that if f is a Riemann integrable function that has k
th
Fourier
coecient equal to a
k
for all k, then

n=1
r
n
n
is bounded for 0 r < 1. This is a contradiction since the sum of the
series is log(1 r).
7. (Exercise 7) Show that the trigonometric series

n2
sin nx
log n
converges for every x, yet it is not the Fourier series of a Riemann inte-
grable function.
The same is true for

sin nx
n

for 0 < < 1, but the case 1/2 < < 1 is


more dicult. See Problem 1.
Solution.
Apply Parsevals identity.(A Riemann integrable function is in L
2
.) Series
converges because the partial sums of

sin nx is bounded; see Exercise


9. Postpone the case 1/2 < < 1 to Problem 1.
8. (Exercise 8) Exercise 6 in Chapter 2 dealt with the sums

n odd 1
1
n
2
=

2
8
and

n=1
1
n
2
=

2
6
.
Similar sums can be derived using the methods of this chapter.
(a) Let f be the function dened on [, ] by f() = [[. Use Parsevals
identity to nd the sums of the following two series:

n=0
1
(2n + 1)
4
and

n=1
1
n
4
.
In fact, they are

4
96
and

4
90
, respectively.
(b) Consider the 2-periodic odd function dened on [0, ] by f() =
( ). Show that

n=0
1
(2n + 1)
6
=

6
960
and

n=1
1
n
6
=

6
945
.
24
Remark. The general expression when k is even for

n=1
1/n
k
in terms
of
k
is given in Problem 4. However, nding a formula for the sum

n=1
1/n
3
, or more generally

n=1
1/n
k
with k odd, is a famous unre-
solved question.
Solution. The Fourier series for f() = [[ is

2

2

n odd 1
e
inx
+e
inx
n
2
.
By Parsevals identity,

2
4
+
4

n odd 1
2
n
4
=
1
2
_

[[
2
d =

2
3
from which we get

n=0
1
(2n + 1)
4
=

4
96
.
The sum

n=1
1
n
4
is obtained from solving for x in the equation
x =

4
96
+
x
16
which yields x =

4
90
.
The Fourier series for the function f() = ( ) is
8

k odd 1
sin k
k
3
= i
4

k odd 1
e
ik
e
ik
k
3
.
By Parsevals identity,
16

n=0
2
(2n + 1)
6
=
1
2
_

2
(1 )
2
d =
1
30

4
,
from which we get

n=0
1
(2n + 1)
6
=

6
960
.
Solving x from
x =
1
64
x +

6
960
we get x =

6
945
.
9. (Exercise 9) Show that for not an integer, the Fourier series of

sin
e
i(x)
25
on [0, 2) is given by

e
inx
n +
.
Apply Parsevals formula to show that

1
(n +)
2
=

2
(sin )
2
.
Solution
Straightforward checking.
10. (Exercise 10) Consider the example of a vibrating string which we analyzed
in Chapter 1. The displacement u(x, t) of the string at time t satises the
wave equation
1
c
2

2
u
t
2
=

2
u
x
2
, c
2
= /.
The string is subject to the initial conditions
u(x, 0) = f(x) and
u
t
(x, 0) = g(x),
where we assume that f C
1
and g is continuous. We dene the total
energy of the string by
E(t) =
1
2

_
L
0
_
u
t
_
2
dx +
1
2

_
L
0
_
u
x
_
2
dx.
The rst term corresponds to the kinetic energy of the string (in analogy
with (1/2)mv
2
, the kinetic energy of a particle of mass m and velocity v),
and the second term corresponds to its potential energy.
Show that the total energy of the string is conserved, in the sense that
E(t) is constant. Therefore,
E(t) = E(0) =
1
2

_
L
0
g(x)
2
dx +
1
2

_
L
0
f

(x)
2
dx.
Solution. We have
E

(t) =
_
L
0
u
t

2
u
t
2
dx +
_
L
0
u
x

2
u
xt
dx
=
_
L
0
u
t

2
u
t
2
dx
_
L
0

2
u

2
x
u
t
dx
= 0
where we have used integration by parts, and that
u
t
(0, t) =
u
t
(L, t) = 0
for all t.
26
11. (Exercise 11) The inequalities of Wirtinger and Poincare establish a rela-
tionship between the norm of a function and that of its derivative.
(a) If f is T-periodic, continuous, and piecewise C
1
with
_
T
0
f(t) dt = 0,
show that
_
T
0
[f(t)[
2
dt
T
2
4
2
_
T
0
[f

(t)[
2
dt,
with equality if and only if f(t) = Asin(2t/T) +Bcos(2t/T).
(b) If f is aa above and g is just C
1
and T-periodic, prove that

_
T
0
f(t)g(t) dt

T
2
4
2
_
T
0
[f(t)[
2
dt
_
T
0
[g

(t)[
2
dt.
(c) For any compact interval [a, b] and any continuously dierentiable func-
tion f with f(a) = f(b) = 0, show that
_
b
a
[f(t)[
2
dt
(b a)
2

2
_
b
a
[f

(t)[
2
dt.
Discuss the case of equality, and prove that the constant (ba)
2
/
2
cannot
be improved.
Solution.
(a) The condition
_
T
0
f(t) dt = 0 implies that

f(0) = 0. The continuity
of f guarantees that

f

(n) =
2in
T

f(n). Indeed, write for


2
T
. Then for
n ,= 0,

f(n) =
1
T
_
T
0
f(t)e
int
dt
=
1
T
(f(0
+
) f(T

))
1
in
+
1
in
1
T
_
T
0
f

(t)e
int
dt
=
T
2in

(n)
Therefore, by Parsevals identity,recalling that

f(0) = 0,
_
T
0
[f(t)[
2
dt = T

|n|>0
[

f(n)[
2
=
T
3
4
2

|n|>0
[

(n)[
2
n
2

T
3
4
2

|n|>0
[

(n)[
2
=
T
3
4
2
1
T
_
T
0
[f

(t)[
2
dt
=
T
2
4
2
_
T
0
[f

(t)[
2
dt
27
From the above inequalities, we see that equality holds if and only if

f(n) = 0 for all n 2. This means that, writing a


n
for

f(n), f(x) =
a
1
e
ix
+a
1
e
ix
which simplies to Asin(x) +Bcos(x).
Remark. It is clear from the proof above that in the absence of the
condition
_
T
0
f(t) dt = 0, the inequality in (a) is

|n|>0
[a
n
[
2

T
2
4
2

|n|>0
[b
n
[
2
where a
n
=

f(n), b
n
=

f

(n).
(b) Let a
n
=

f(n), b
n
= g(n), c
n
=

g

(n). Then

_
T
0
f(t)g(t) dt

2
= T[

|n|0
a
n
b
n
[
2
= T[

|n|>0
a
n
b
n
[
2
(T

|n|>0
[a
n
[
2
)(

|n|>0
[b
n
[
2
)

_
T
0
[f(t)[
2
dt
T
2
4
2
_
T
0
[g

(t)[ dt
(c) Extend f to a function on [a, 2b a] such that f(b + h) = f(b
h) for 0 h b a and then extend it so that it is T = 2(b a)-
periodic. It is easy to see that now f(b +h) = f(b h) and f

(b +h) =
f

(b h) for all h. Then


_
T
0
f(x) dx =
_
2ba
a
f(x) dx = 0,
_
T
0
[f(x)[ dx =
_
2ba
a
[f(x)[ dx = 2
_
b
a
[f(x)[ dx, and
_
T
0
[f

(x)[ dx = 2
_
b
a
[f

(x)[ dx. Check


that f so extended is also C
1
on R. Therefore by (a), we have
_
b
a
[f(x)[ dx
=
1
2
_
T
0
[f(x)[ dx

1
2
2
2
(b a)
2
4
2
_
T
0
[f

(x)[ dx
=
(b a)
2

2
1
2
_
T
0
[f

(x)[ dx
=
(b a)
2

2
_
b
a
[f

(x)[ dx
For the function f(x) that we dene, its translation g(x) = f(x +b) is an
odd function. So g(n) = e
inb
f(n), and the Fourier series for g is of the
28
form

n=1
A
n
sin
2nx
2(b a)
=

n=1
A
n
sin
nx
b a
.
Thus the Fourier series for f is of the form

n=1
A
n
sin
n(x b)
b a
.
According the inequality in (c) is an equality i f(x) is of the form
Asin
(xb)
ba
. (The book says Asin
(xa)
ba
; this is equivalent since sin
(xb)
ba
=
sin
(xa)
ba
).
12. (Exercise 12) Prove that
_

0
sin x
x
dx =

2
.
Proof.
We have
_

D
N
(t) dt = 2. So
_

sin(N + 1/2)x
sin(x/2)
dx = 2
Write csc(x/2) as csc(x/2) 2/x + 2/x. lim
x0
csc(x/2) 2/x = 0, so it
has a removable discontinuity at 0 on the interval [, ]. By Lebesgue-
Riemann lemma we get
_

2 sin(N + 1/2)x
x
dx 2 as N .
This yields
_

0
sin(N + 1/2)x
x
dx

2
as N .
By change of variable, we get
_
(N+1/2)
0
sin x
x
dx

2
as N .
Since
_
(N+1/2)
N
sin x
x
dx 0 as N , (use Mean-Value Theorem), we
get
_
N
0
sin x
x
dx

2
as N .
By MVT,
_
(N+1)
N
[ sin x[
x
dx
1
N
,
so for any t > there exists N > 0 such that

_
t
N
sin x
x
dx

1
N
.
29
It follows that
_

0
sin x
x
dx = lim
N
_
N
0
sin x
x
dx =

2
.
13. (Exercise 13) Suppose that f is periodic and of class C
k
. Show that

f(n) = o(1/[n[
k
),
that is, [n[
k
f(n) goes to 0 as [n[ . This is an improvement over
Exercise 10 in Chapter 2.
Solution.
We have (see p.43)
2

f(n) =
1
(in)
k
_
2
0
f
(k)
()e
in
d.
So
2(in)
k

f(n) =
_
2
0
f
(k)
()e
in
d.
Now use Lebesgue-Riemann lemma.
14. (Exercise 14) Prove that the Fourier series of a continuously dierentiable
function f on the circle is absolutely convergent.
Proof.
From

f(n) =
1
in

(n), we get, by Cauchy-Schwarz inequality and Parse-


vals identity,

n=
[

f(n)[
=

n=
1
[n[
[

(n)[
(2

2
6
)
1/2
(

n=
[

(n)[
2
)
1/2
= (2

2
6
)
1/2
1
2
_

[f

(t)[
2
dt < .
15. (Add in) Prove that the Fourier series of a 2 periodic absolutely contin-
uous function whose derivative (exists a.e.) in [0, 2] is square integrable
(in particular, Riemann integrable), is absolutely convergent. (Note that
Exercise 16 below shows that Lipschitz condition alone is enough. But
derivative of a Lipschitz function is bounded. And Lipschitz functions
are precisely functions representable as integral of a bounded measurable
function.)
Proof. Use the proof in Exercise 14. Note that the integration by parts
formula is valid for absolutely continuous functions.
30
16. (Exercise 15) Let f be a 2-periodic and Riemann integrable on [, ].
(a) Show that

f(n) =
1
2
_

f(x +

n
)e
inx
dx
hence

f(n) =
1
4
_

[f(x) f(x +

n
)]e
inx
dx.
(b) Now assume that f satises a Holder condition of order , namely
[f(x +h) f(x)[ C[h[

for some 0 < 1, some C > 0, and all x, h. Use part (a) to show that

f(n) = O(1/[n[

).
(c) Prove that the above result cannot be improved by showing that the
function
f(x) =

k=0
2
k
e
i2
k
x
,
where 0 < < 1, satises
[f(x +h) f(x)[ C[h[

,
and

f(N) = 1/N

whenever N = 2
k
.
Solution.
Note that for any real x, [1 e
ix
[ [x[ (to prove, just note that LHS is
2[ sin(x/2)[ and [ sin a[ [a[ for all real number a).
(a).
1
2
_

f(x +

n
)e
inx
dx
=
1
2
_
+

n
+

n
f(u)e
inu
e
i
du
=
1
2
_

f(u)e
inu
du
=

f(n)
(b)
[

f(n)[

1
4
_

[f(x) f(x +

n
)[ dx

1
4
2C

[n[

=
C
1
[n[

31
(c).
[f(x +h) f(x)[
= [

k=0
2
k
e
i2
k
(x+h)

k=0
2
k
e
i2
k
x
[

k=0
2
k
[e
i2
k
h
1[

2
k
1/|h|
2
k
2
k
[h[ +

2
k
>1/|h|
2
k
2
The second sum is easily seen to be less than 2[h[

. The rst sum is 0 if


[h[ > 1. So assume [h[ 1. Let l be the unique nonnegative integer such
that 2
l1
< [h[ 2
l
. Then the rst sum is
l

k=0
(2
k
[h[)
1
[h[

[h[

k=0
(2
kl
)
1

1
1 2
1
[h[

.
Since the series converges uniformly, the coecient of e
inx
is

f(n) for all
n.
17. (Exercise 16) Let f be a 2-periodic function which satises a Lipschitz
condition with constant K; that is
[f(x) f(y)[ K[x y[ for all x, y.
This is simply the Holder condition with = 1, so by the previous exercise,
we see that

f(n) = O(1/[n[). Since the harmonic series

1/n diverges, we
cannot say anything (yet) about the absolute convergence of the Fourier
series of f. The outline below actually proves that the Fourier series of f
converges absolutely and uniformly.
(a) For every positive h we dene g
h
(x) = f(x+h) f(xh). Prove that
1
2
_
2
0
[g
h
(x)[
2
dx =

n=
4[ sin nh[
2
[

f(n)[
2
,
and show that

n=
[ sin nh[
2
[

f(n)[
2
K
2
h
2
.
(b) Let p be a positive integer. By choosing h = /2
p+1
, show that

2
p1
<|n|2
p
[

f(n)[
2

K
2

2
2
2p+1
.
(c) Estimate

2
p1
<|n|2
p [

f(n)[, and conclude that the Fourier series of


f converges absolutely, hence uniformly.
32
(d) In fact, modify the argument slightly to prove Bernsteins theorem: If
f satises a Holder condition of order > 1/2, then the Fourier series of
f converges absolutely.
Solution.
(a). The Fourier coecients of the translated function f(x+h) is e
inh
f(n).
So
g
h
(n) = (e
inh
e
inh
)

f(n) = 2i sin nh

f(n).
The rst equation in (a) follows from Parsevals identity. Since [g
h
(x)[
K[x +h (x h)] = 2K[h[, the second inequality in (a) follows.
(b). h = /2
p+1
and 2
p1
< [n[ 2
p
imply /4 < [n[h /2, hence
[ sin nh[
2
1/2. Thus
1
2

2
p1
<|n|2
p
[

f(n)[
2

n=
[ sin nh[
2
[

f(n)[
2
K
2
h
2
=
K
2

2
2
2(p+1)
from which it follows

2
p1
<|n|2
p
[

f(n)[
2

K
2

2
2
2p+1
.
(c). By Cauchy-Schwarz inequality, we have

2
p1
<|n|2
p
[

f(n)[
(

2
p1
<|n|2
p
1
2
)
1/2
(

2
p1
<|n|2
p
[

f(n)[
2
)
1/2
(2
p
2
p1
)
1/2
K
2
p+1/2

K
(

2)
p
Therefore

1|n|<
[

f(n)[

1p<
K
(

2)
p
<
(d). Modications are given below:
The Fourier coecients of the translated function f(x+h) is e
inh
f(n). So
g
h
(n) = (e
inh
e
inh
)

f(n) = 2i sin nh

f(n).
The rst equation in (a) follows from Parsevals identity. Since [g
h
(x)[
K[x +h (x h)[

= 2

K[h[

, the second inequality in (a) becomes

n=
[ sin nh[
2
[

f(n)[
2
2
2(1)
K
2
[h[
2
.
33
h = /2
p+1
and 2
p1
< [n[ 2
p
imply /4 < [n[h /2, hence
[ sin nh[
2
1/2. Thus
1
2

2
p1
<|n|2
p
[

f(n)[
2

n=
[ sin nh[
2
[

f(n)[
2
2
2(1)
K
2
[h[
2
=
K
2

2
2
2(p+1)
from which it follows

2
p1
<|n|2
p
[

f(n)[
2

K
2

2
2
2p+1
.
By Cauchy-Schwarz inequality, we have

2
p1
<|n|2
p
[

f(n)[
(

2
p1
<|n|2
p
1
2
)
1/2
(

2
p1
<|n|2
p
[

f(n)[
2
)
1/2
(2
p
2
p1
)
1/2
K

2
p+1/2

K
(2
1/2
)
p
Therefore, if > 1/2, then

1|n|<
[

f(n)[

1p<
K
(2
1/2
)
p
<
Remark. The condition > 1/2 is sharp. See Katznelsons book, p. 32.
18. (Exercise 17) If f is a bounded monotonic function on [, ], then

f(n) = O(1/[n[).
19. (Exercise 18) Here are a few things we have learned about the decay of
Fourier coecients:
(a) if f is of class C
k
, then

f(n) = o(1/[n[
k
);
(b) if f is Lipschitz, then

f(n) = O(1/[n[);
(c) if f is monotonic, then

f(n) = O(1/[n[);
(d) if f satises a Holder condition with exponent where 0 < < 1,
then

f(n) = O(1/[n[

);
(e) if f is merely Riemann integrable, then

f(n)[
2
< and therefore

f(n) = o(1).
Nevertheless, show that the Fourier coecients of a continuous function
can tend to 0 arbitrarily slowly by proving that for every sequence of
nonnegative real numbers
k
converging to 0, there exists a continuous
34
function f such that [

f(n)[
n
for innitely many values of n.
Solution.
Choose a subsequence
n
k
such that

k

n
k
< , e.g.,
n
k
< 1/2
k
.
Then the function

k=1

n
k
e
in
k
x
, is such a function, because it is abso-
lutely ( and hence uniformly) convergent.
20. (Exercise 19) Give another proof that the sum

0<|n|N
e
inx
/n is uni-
formly bounded in N and x [, ] by using the fact that
1
2i

0<|n|N
e
inx
n
=
N

n=1
sin nx
n
=
1
2
_
x
0
(D
N
(t) 1) dt,
where D
N
is the Dirichlet kernel. Now use the fact that
_

0
sin t
t
dt <
which was proved in Exercise 12.
21. (Exercise 20) Let f(x) denote the sawtooth function dened by f(x) =
(x)/2 on the interval (0, 2) with f(0) = 0 and extended by periodicity
to all of R. The Fourier series of f is
f(x)
1
2i

n=0
e
i
nx
n
=

n=1
sin nx
n
,
and f has a jump discontinuity at the origin with
f(0
+
) =

2
, f(0

) =

2
, and hence f(0
+
) f(0

) = .
Show that
lim
N
max
0<x/N
S
N
(f)(x)

2
=
_

0
sin t
t
dt

2
0.08949,
which is roughly 9% of the jump . This result is a manifestation of
Gibbss phenomenon which states that near a jump discontinuity, the
Fourier series of a function overshoots (or undershoots) it by approxi-
mately 9% of the jump.
22. (Problem 1) For each 0 < < 1 the series

n=1
sin nx
n

converges for every x but is not the Fourier series of a Riemann integrable
function.
(a) If the conjugate Dirichlet kernel is dened by

D
N
(x) =

|n|N
sign(n)e
inx
where sign(n) =
_
_
_
1 if n > 0
0 if n = 0
1 if n < 0,
35
then show that

D
N
(x) = i
cos(x/2) cos((N + 1/2)x)
sin(x/2)
,
and
_

D
N
(x)[ dx c log N, for N 2
(b) As a result, if f is Riemann integrable, then
(f

D)(0) = O(log N).
(c) In the present case, this leads to

n=1
1
n

= O(log N),
which is a contradiction.
Solution.
Let = e
ix
.

D
N
(x)
=
n

k=1

k=1

k
=
1
n
1

1
1
n
1
1
=
1
n
1
+
1
n
1
=
+ 1
n+1

n
1
=

1/2
+
1/2

n+1/2

n1/2

1/2

1/2
=
cos(x/2) cos((N + 1/2)x)
sin(x/2)
i
Note that by trigonometric identity, this is
2 sin(
N+1
2
x) sin(
N
2
x)
sin(x/2)
i.
Note that t/ sin t /2 for t [/2, /2], so for x [, ],
[

D
N
(x)[ 4
[ sin(
N
2
x)[
x
36
Then
_

D
N
()[ d
8
_

0
[ sin(N/2)[

d
= 8
_
(N/2)
0
[ sin [

d
= 8
N1

k=0
_
(k+1)/2
k/2
[ sin [

d
= 8
N1

k=1
_
(k+1)/2
k/2
[ sin [

d +
_
/2
0
sin

d
8
2

N1

k=1
1
k
_
(k+1)/2
k/2
[ sin [ d +
_
/2
0
sin

d
=
16

N1

k=1
1
k
+
_
/2
0
sin

d
=
16

N1

k=2
1
k
+
16

+
_
/2
0
sin

16

log N +c
1
c log N
where c is chosen such that c/2
16

and (c/2) log 2 c


1
, so that
c log N = c/2 log N +c/2 log N
16

log N +c
1
for all N 2.
(b),(c). If f is such a Riemann integrable function, then

f(n) = 1/n

,

f(n) =
37
1/n

for n > 0, and



f(0) = 0. So
(f

D
N
)(0) =
1
2
_

f(t)

D
N
(0 t) dt
=
1
2
_

f(t)

D
N
(t) dt
=
1
2
_

f(t)

|n|N
sign(n)e
int
dt
=

|n|N
sign(n)
1
2
_

f(t)e
int
dt
=

|n|N
sign(n)

f(n)
= 2
N

n=1
1
n

Since f(t) is bounded, we also have by (a),


[
_

f(t)

D
N
(t) dt[ M
_

D
N
(t)[ dt = O(log N).
Thus

N
n=1
1
n

= O(log N); this is impossible (since

N
n=1
1
n

grows like
N
1
, see my ApostolStudyNote, p. 25).
23. (Problem 2)
24. (Problem 3) Let be a complex number not equal to an integer.
(a) Calculate the Fourier series of the 2-periodic function dened on
[, ] by f(x) = cos(x).
(b) Prove the following formulas due to Euler:

n=1
1
n
2

2
=
1
2
2


2tan()
.
For all u C Z,
cot u =
1
u
+ 2

n=1
u
u
2
n
2

2
.
(c) Show that for all C Z we have

sin()
= 1 + 2
2

n=1
(1)
n1
n
2

2
.
(d) For all 0 < < 1, show that
_

0
t
1
t + 1
dt =

sin()
.
38
Solution.
From Maple, for all n,

f(n) = (1)
n+1
sin()
(n
2

2
)
.
It follows that for x ,
cos(x) =
sin()

+
2sin()

n=1
(1)
n+1
cos nx
n
2

2
.
Evaluating at x = yields the rst equation in part (b). The second
equation follows directly from the rst by substituting u for .
(c) follows by substituting x = 0 in the Fourier series.
(d). Split the integral as
_
1
0
+
_

1
and change variables t = 1/u in the
second integral, one gets
_

0
t
1
t + 1
dt =
_
1
0
t
1
t + 1
dt +
_
1
0
t
(1)1
t + 1
dt.
Claim: For 0 < < 1,
_
1
0
t
1
t + 1
dt =

k=0
(1)
k
1
k +
.
Proof.
For each 0 < r < 1, the series t
1

k=0
(1)
k
t
k
converges to
t
1
t+1
uni-
formly on [0, r]. So
_
r
0
t
1
t + 1
dt =

k=0
(1)
k
r
+k
k +
Since
_
1
0
t
1
t + 1
dt
= lim
r1

_
r
0
t
1
t + 1
dt
= lim
r1

k=0
(1)
k
r
+k
k +
= lim
r1

k=0
(1)
k
r
k
k +
The series

k=0
(1)
k 1
k+
is Abel summable and hence convergent to
_
1
0
t
1
t+1
dt, by Littlewoods theorem since (1)
k
1/(k+) = O(1/k). Q.E.D.
39
Therefore
_
1
0
t
1
t + 1
dt +
_
1
0
t
(1)1
t + 1
dt
=

k=0
(1)
k
k + 1
+

k=0
(1)
k
k +
=

k=1
(1)
k1
k
+
1

k=1
(1)
k
k +
=
1

+ 2

k=1
(1)
k1
k
2

2
=

sin()
by (c).
25. (Problem 4) In this problem, we nd the formula for the sum of the series

n=1
1
n
k
where k is any even integer. These numbers are expressed in terms of the
Bernoulli numbers; the related Bernoulli polynomials are discussed in the
next problem.
Dene the Bernoulli numbers B
n
by the formula
z
e
z
1
=

n=0
B
n
n!
z
n
.
(a) Show that B
0
= 1, B
1
= 1/2, B
2
= 1/6, B
3
= 0, B
4
= 1/30, and
B
5
= 0.
(b) Show that for n 1 we have
B
n
=
1
n + 1
n1

k=0
_
n + 1
k
_
B
k
.
(c) By writing
z
e
z
1
= 1
z
2
+

n=2
B
n
n!
z
n
,
show that B
n
= 0 if n is odd and > 1. Also prove that
z cot z = 1 +

n=1
2
2n
B
2n
(2n)!
z
2n
.
40
(d) The zeta function is dened by
(s) =

n=1
1
n
s
, for all s > 1.
Deduce from the result in (c), and the expression for the cotangent func-
tion obtained in the previous problem, that
xcot x = 1 2

m=1
(2m)

2m
x
2m
.
(e) Conclude that
2(2m) = (1)
m+1
(2)
2m
(2m)!
B
2m
.
Solution.
(a) Get these numbers directly from (1 + z/2! + z
2
/3! + )
1
by long
division.
(b).
1 =
z
e
z
1
e
z
1
z
=
_

n=0
B
n
n!
z
n
__

n=0
1
(n + 1)!
z
n
_
=

n=0
_
n

k=0
B
k
k!
1
(n k + 1)!
_
t
n
from which it follows that B
0
= 1, and for n 1,
n

k=0
B
k
k!
1
(n k + 1)!
= 0,
i.e.
B
n
n!
=
n1

k=0
B
k
k!
1
(n k + 1)!
=
1
(n + 1)!
n1

k=0
_
n + 1
k
_
B
k
(c). Since B
0
= 1, B
1
=
1
2
, we have
z
e
z
1
= 1
z
2
+

n=2
B
n
n!
z
n
.
41
Now,
z
e
z
1
+
z
2
= z
_
1
e
z
1
+
1
2
_
=
z
2
_
e
z
+ 1
e
z
1
_
=
z
2
_
e
z/2
+e
z/2
e
z/2
e
z/2
_
=
z
2
coth
z
2
is an even function, so B
2n+1
= 0 for all n 1. Note that
coth(iz) =
e
iz/2
+e
iz/2
e
iz/2
e
iz/2
=
2 cos z
2i sin z
so cot z = i coth(iz). Therefore
z cot z = iz coth(iz) = 1 +

n=1
B
2n
(2n)!
(i2z)
2n
=

n=0
(1)
n
2
2n
B
2n
(2n)!
z
2n
From Problem 3 (b), we have for 0 < x < ,writing r
n
for
x
n
,
xcot x = 1 2

n=1
x
2
n
2

2
x
2
= 1 2

n=1
r
2
n
1 r
2
n
= 1 2

n=1

m=1
r
2m
n
= 1 2

m=1
_

m=1
1
n
2m
_
x
2m

2m
= 1 2

m=1
(2m)

2m
x
2m
Comparing coecients of xcot x in both formulas, we get (e).
Remark. Ross Tang has discovered explicit formulas for Bernoulli num-
bers and Euler numbers in
http://www.voofie.com/content/117/an-explicit-formula-for-the-euler-zigzag-numbers
-updown-numbers-from-power-series/#Explicit_Formula_for_Euler_number
42
His formula is
B
2n
=
2n
2
2n
4
2n
2n

k=1
k

j=0
_
k
j
_
(1)
j
(k 2j)
2n
2
k
i
k
k
,
E
2n
= i
2n+1

k=1
k

j=0
_
k
j
_
(1)
j
(k 2j)
2n+1
2
k
i
k
k
.
Update: Simpler formulas are already in
http://en.wikipedia.org/wiki/Bernoulli_number
Perhaps it is a good project to try to nd explicit formulas for
Bernoulli polynomials.
From the above solution, we get the following series expansion for cot x
and coth x:
cot z =
1
z
+

n=1
(1)
n
2
2n
B
2n
(2n)!
z
2n1
,
coth z =

n=1
2
2n
B
2n
(2n)!
z
2n1
.
I have yet to gure out the proofs of the next four formulas:
tan z =

n=1
(1)
n1
4
n
(4
n
1)B
2n
(2n)!
z
2n1
= z +
z
3
3
+
2z
5
15
+
17z
7
315
+ ,
tanh z =

n=1
4
n
(4
n
1)B
2n
(2n)!
z
2n1
= z
z
3
3
+
2z
5
15

17z
7
315
+ ,
sec z =

n=0
(1)
n
E
2n
(2n)!
z
2n
,
sech z =

n=0
E
2n
(2n)!
z
2n
,
where E
2n
are the Eulers numbers. (The generating function for E
2n
is
sech z =
2
e
z
+e
z
).
Update: First note that
2 coth(2z) coth z = tanh z
from which we get the series for tanh z. Then use tan z = i tanh(iz) to
get tan z series. Other series that use Bernoulli numbers are:
z
sin z
,
z
sinh z
, log
sin z
z
, log(cos z), log
tan z
z
, ,
43
but I have not looked into this assertion.
Atkinson (American Math Monthly, vol. 93, no. 5,1986, p. 387-; this
paper in pdf form is in my computer under the name TangentSeries.pdf)
has discovered that the above series for tangent and secant can be read
o from the sides of the following triangle (only seven rows are shown):
1
0 1
1 1 0
0 1 2 2
5 5 4 2 0
0 5 10 14 16 16
61 61 56 46 32 16 0

The eighth row is 0, 61, 122, 178, 224, 256, 272, 272.
26. (Problem 5) Dene the Bernoulli polynomials B
n
(x) by the formula
ze
xz
e
z
1
=

n=0
B
n
(x)
n!
z
n
.
(a) The functions B
n
(x) are polynomials in x and
B
n
(x) =
n

k=0
_
n
k
_
B
k
x
nk
.
Show that B
0
(x) = 1, B
1
(x) = x1/2, B
2
(x) = x
2
x+1/6, and B
3
(x) =
x
3

3
2
x
2
+
1
2
x.
(b) If n 1, then
B
n
(x + 1) B
n
(x) = nx
n1
,
and if n 2, then
B
n
(0) = B
n
(1) = B
n
.
(c) Dene S
m
(n) = 1
m
+ 2
m
+ + (n 1)
m
. Show that
(m+ 1)S
m
(n) = B
m+1
(n) B
m+1
.
(d) Prove that the Bernoulli polynomials are the only polynomials that
satisfy (i) B
0
(x) = 1, (ii) B

n
(x) = nB
n1
(x) for n 1, (iii)
_
1
0
B
n
(x) dx =
0 for n 1, and show that from (b) one obtains
_
x+1
m
B
n
(t) dt = x
n
.
(e) Calculate the Fourier series of B
1
(x) to conclude that for 0 < x < 1
we have
B
1
(x) = x
1
2
=
1

k=1
sin(2kx)
k
.
44
Integrate and conclude that
B
2n
(x) = (1)
n+1
2(2n)!
(2)
2n

k=1
cos(2kx)
k
2n
,
B
2n+1
(x) = (1)
n+1
2(2n + 1)!
(2)
2n+1

k=1
sin(2kx)
k
2n+1
.
Finally, show that for 0 < x < 1,
B
n
(x) =
n!
(2i)
n

k=0
e
2ikx
k
n
.
We observe that the Bernoulli polynomials are, up to normalization (i.e.
requiring
_
1
0
B
n
(x) dx = 0), successive integrals (antiderivatives) of the
sawtooth function x 1/2, 0 < x < 1.
Solution.
ze
xz
e
z
1
=
z
e
z
1
e
xz
=

n=0
B
n
n!
z
n

n=0
x
n
n!
z
n
=

n=0
_
n

k=0
B
k
k!
x
nk
(n k)!
_
z
n
=

n=0
1
n!
_
n

k=0
_
n
k
_
B
k
x
nk
_
z
n
Therefore
B
n
(x) =
n

k=0
_
n
k
_
B
k
x
nk
.
ze
(x+1)z
e
z
1

ze
xz
e
z
1
=
ze
xz
e
z
1
(e
z
1)
= ze
xz
=

n=0
x
n
n!
z
n+1
=

n=1
x
n1
(n 1)!
z
n
45
Therefore comparing coecients of z
n
, n 1 we get
B
n
(x + 1)
n!

B
n
(x)
n!
=
x
n1
(n 1)!
and hence
B
n
(x + 1) B
n
(x) = nx
n1
,
So for n 2, B
n
(1) B
n
(0) = n0
n1
= 0,i.e. B
n
(1) = B
n
(0) = B
n
.
(Note that by denition of B
n
(x), we have B
n
(0) = B
n
.)
For m 1,
(m+ 1)
n1

k=1
k
m
=
n1

k=1
B
m+1
(k + 1) B
m+1
(k)
= B
m+1
(n) B
m+1
(1)
= B
m+1
(n) B
m+1
Note that this yields a formula for the sum
1
m
+ 2
m
+ +n
m
=
1
m+ 1
(B
m+1
(n + 1) B
m+1
).
Write F(x, z) for
ze
xz
e
z
1
. Since

x
F(x, z) = zF(x, z)
, comparing coecient of z
n
on both sides, we get
B

n
(x)
n!
=
B
n1
(x)
(n 1)!
,
hence B

n
(x) = nB
n1
(x). Next,
_
1
0
B
n
(x) dx =
1
n + 1
_
1
0
B

n+1
(x) dx =
1
n + 1
(B
n+1
(1) B
n+1
(0)) = 0.
From (b),
_
x+1
x
B
n
(t) dt =
1
n + 1
_
x+1
x
B

n+1
(t) dt =
1
n + 1
(B
n+1
(x+1)B
n+1
(x)) = x
n
.
From Maple, one gets
B
1
(x) = x
1
2
=
1

k=1
sin(2kx)
k
.
46
B
2
(x) is an antiderivative of 2B
1
(x), so
B
2
(x) =
2

k=1
cos(2kx)
2k
2
+C
But
_
1
0
B
2
(x) dx = 0 and
_
1
0
cos(2kx) dx = 0, so C = 0, and
B
2
(x) =
2
2
2

k=1
cos(2kx)
2k
2
=
2 2!
(2)
2

k=1
cos(2kx)
k
2
.
Next, arguing the same manner, B
3
(x) is an antiderivative of 3B
2
(x), and
B
3
(x) =
2 3!
(2)
3

k=1
sin(2kx)
k
3
.
And then
B
4
(x) =
2 4!
(2)
4

k=1
cos(2kx)
k
4
.
and so on, and we get
B
2n
(x) = (1)
n+1
2(2n)!
(2)
2n

k=1
cos(2kx)
k
2n
,
B
2n+1
(x) = (1)
n+1
2(2n + 1)!
(2)
2n+1

k=1
sin(2kx)
k
2n+1
.
Writing sin, cos in terms of e powers, and noting that i
2m
= (1)
m
, we
get the last formula
B
n
(x) =
n!
(2i)
n

k=0
e
2ikx
k
n
.
Chapter 4
1. (Problem 5) Let f be a Riemann integrable function on the interval [, ].
We dene the generalized delayed means of the Fourier series of f by

N,K
=
S
N
+ +S
N+K1
K
.
Note that in particular

0,N
=
N
,
N,1
= S
N
and
N,N
=
N
,
47
where
N
are the specic delayed means used in Section 3.
(a) Show that

N,K
=
1
K
((N +K)
N+K
N
N
),
and

N,K
= S
N
+

N+1|j|N+K1
_
1
[j[ N
K
_

f(j)e
ij
.
From this last expression for
N,K
conclude that
[
N,K
S
M
[

N+1|j|N+K1
[

f(j)[
for all N M < N +K.
(b) Use one of the above formulas and Fejers theorem to show that with
N = kn and K = n, then

kn,n
(f)() f() as n
whenever f is continuous at , and also

kn,n
(f)()
f(
+
) +f(

)
2
as n
at a jump discontinuity (refer to the preceding chapters and their exer-
cises for the appropriate denitions and results). In the case when f is
continuous on [, ], show that
kn,n
(f) f uniformly as n .
(c) Using part (a), show that if

f(j) = O(1/[j[) and kn m < (k + 1)n,
we get
[
kn,n
S
m
[
C
k
for some constant C > 0.
(d) Suppose that

f(j) = O(1/[j[). Prove that if f is continuous at then
S
N
(f)() f() as N ,
and if f has a jump discontinuity at then
S
N
(f)()
f(
+
) +f(

)
2
as N .
Also, show that if f is continuous on [, ], then S
N
(f) f uniformly.
(e) The above arguments show if

c
n
is Ces`aro summable to s and c
n
=
O(1/n), then

c
n
converges to s. This is a weak version of Littlewoods
theorem (Problem 3, Chapter 2).
Solution.
48
(a).
N,K
=
1
K
((N + K)
N+K
N
N
) is straightforward. Write a
j
for

f(j)e
ij
. We have

N,K
=
1
K
[S
N
+ (S
N
+a
N+1
+a
N1
) + + (S
N
+

N+1|j|N+K1
a
j
)]
= S
N
+
1
K
K1

l=1

N+1|j|N+l
a
j
= S
N
+
1
K

N+1|j|N+K1
a
j
K1

l=|j|N
1
= S
N
+
1
K

N+1|j|N+K1
(K +N [j[)a
j
= S
N
+

N+1|j|N+K1
_
1
[j[ N
K
_
a
j
For N M < N + K,
N,K
S
M
=

N+1|j|N+K1
b
j
a
j
, where b
j
is
either 1
|j|N
K
or
|j|N
K
, hence [
N,K
S
M
[

N+1|j|N+K1
[a
j
[.
(b). We have

kn,n
=
1
n
[(k + 1)n
(k+1)n
kn
kn
] = (k + 1)
(k+1)n
k
kn
Let A be either f() or
f(
+
)+f(

)
2
. Since
(k+1)n
,
kn
approach the
same value A, as n , we see that
kn,n
A. Uniform convergence
statement is also clear from
kn,n
= (k +1)
(k+1)n
k
kn
. (I think I have
to elaborate this last sentence.)
(c). Suppose [

f(j)[ = [a
j
[ C
1
/[j[ for some constant C
1
> 0. From the
last part of (a), for kn m < (k + 1)n,
[
kn,n
S
m
[
C
1

kn+1|j|(k+1)n1
1
[j[
2C
1
n 1
kn + 1
2C
1
1
k
=
C
k
(d). Let > 0. Fix a positive integer k such that
C
k
< . Choose N such
that
[
kn,n
A[ <
49
for all n N. Then for all m kN, we have kn m < (k +1)n for some
n N, and
[S
m
A[ [S
m

kn,n
[ +[
kn,n
A[ < 2.
This proves that S
m
A as m .
(e)Note that except for the uniform convergence part, the only properties
of the series

j
a
j
that we use in the above proof are that it is Cesaro
summable to A and that a
j
= O(1/[j[).
Chapter 5 The Fourier Transform on R
1. (Exercise 1) Corollary 2.3 in Chapter 2 leads to the following simplied
version of the Fourier inversion formula. Suppose f is a continuous func-
tion supported on an interval [M, M], whose Fourier transform

f is of
moderate decrease.
(a) Fix L with L/2 > M, and show that f(x) =

a
n
(L)e
2inx/L
where
a
n
(L) =
1
L
_
L/2
L/2
f(x)e
2inx/L
dx =
1
L

f(n/L).
Alternatively, we may write f(x) =

n=

f(n)e
2inx
with = 1/L.
(b) Prove that if F is continuous and of moderate decrease, then
_

F() d = lim
0,>0

n=
F(n).
(c) Conclude that f(x) =
_

f()e
2ix
d.
Solution.
(a)

n=
a
n
(L)e
2inx/L
=
1
L

n=

f(
n
L
)e
2inx/L
Let B, C be the constants dened as in Remark 1 below. Since [

f(
n
L
)[
B
n
2
, the above series is absolutely convergent. Hence the series converges
uniformly (and absolutely) to f(x) by Corollary 2.3 in Chapter 2.
(b). Given > 0, choose N > C such that
_
|x|>N
[F(x)[ dx < /4 and

|n|>N
B
n
2
< /4. Choose 0 <
1
< 1 such that for all 0 < <
1
,
[
_
N
N
F(x) dx

|n|N/
F(n)[ <

2
.
This is possible because

|n|N/
F(n) is almost a Riemann sum. (Fill
in the details.) Since
[

|n|>N/
F(n)[

|n|>N
[F(n)[

|n|>N
B
n
2
50
the result follows.
(c). Apply (b) to F() =

f()e
2ix
and use (a).
Remark 1. From the books denition, a function f dened on R is said
to be of moderate decrease if f is continuous and there exists a constant
A > 0 so that
[f(x)[
A
1 +x
2
for all x R.
Note that this is equivalent to saying that f is continuous and there are
positive constants B, C such that
[f(x)[
B
x
2
for all [x[ C.
Proof. Suppose f satises the rst condition. Since for all x ,= 0
A
1 +x
2
=
A
1 +x
2
/2 +x
2
/2

2A
x
2
we can choose B = 2A, C = 1 in the second statement.
Now suppose f satises the second condition. Let D = max1, C. Then
B
x
2
=
2B
x
2
+x
2

2B
1 +x
2
for [x[ D.
Let M = max(1+x
2
)[f(x)[ : [x[ D. Then we can let A = max2B, M
in the rst condition.
Remark 2. Let f(x) = 1, 1 x 1, zero elsewhere. Then the Fourier
transform of f is
sin(2

, which is not integrable. Modifying f, making it


continuous, we let f
n
be the even extension of the following function
g(x) = 1 for 0 x
n 1
n
, and n(x 1) for
n 1
n
x 1
and zero for x 1. The Fourier transform of f
n
, n 1, is
n(1 2 cos
2
() + cos(
n1
n
2))
2
2

2
which is of moderate decrease. A continuous with compact support such
it Fourier transform is not of moderate decrease is in Exercise 3. It
remains to see an example of continuous function of compact
support such that its Fourier transform is not integrable. From
sci.math, cronopio said Terry Tao seems to say that the function (1
log(1 [x[))
a
, [x[ < 1, zero elsewhere, (0 < a < 1 is a xed number) is
an example. See http://mathoverow.net/questions/43550/is-the-fourier-
transform-of-1-1-log1-x2-supported-in-1-1-integrable
Existence of such function also follows from closed graph theorem, see the
link in the above link.
51
2. (Exercise 2) Let f and g be the functions dened by
f(x) =
[1,1]
(x) =
_
1 if [x[ 1,
0 otherwise,
and g(x) =
_
1 [x[ if [x[ 1,
0 otherwise.
Although f is not continuous, the integral dening its Fourier transform
still make sense. Show that

f() =
sin 2

and g() =
_
sin

_
2
,
with the understanding that

f(0) = 2 and g(0) = 1.
Solution. Check with Maple and found to agree.
3. (Exercise 3) The following exercise illustrates the principle that the decay
of

f is related to the continuity properties of f.
(a) Suppose that f is a function of moderate decrease on R whose Fourier
transform

f is continuous and satises

f() = O
_
1
[[
1+
_
as [[
for some 0 < < 1. Prove that f satises a Holder condition of order ,
that is, that
[f(x +h) f(x)[ M[h[

for some M > 0 and all x, h R.


(b) Let f be a continuous function on R which vanishes for [x[ 1, with
f(0) = 0, and which is equal to 1/ log(1/[x[) for all x in a neighborhood
of the origin. Prove that

f is not of moderate decrease. In fact, there is
no > 0 so that

f() = O(1/[[
1+
) as [[ .
Solution.
By remarks in section 1.7, since both f and

f are of moderate decrease,
the inverse Fourier transform of

f is f. Thus
f(x +h) f(x) =
_

f()e
2ix
(e
2ih
1) d.
The condition on

f is equivalent to
[

f()[
A
1 +[[
1+
for all R
52
for some A > 0. Thus (note that [e
2ih
1[ = 2[ sin(h)[)

f(x +h) f(x)


h

1
[h[

A[e
2ih
1[
1 +[[
1+
d

4A
[h[

_

0
[ sin(h)[
1 +
1+
d
=
4A

_

0
[ sin(u)[
[h[
1+
+u
1+
du

4A

_
_
1
0
[
sin(u)
u
[
u

du +
_

1
1
u
1+
du
_

4A

__
1
0
1
u

du +
_

1
1
u
1+
du
_
<
(b). We have
[f(h) f(0)[
[h[

=
1
[h[

log [h[

as h 0 for any xed > 0. So by (a),

f is not of moderate decrease.
4. (Exercise 4) Examples of compactly supported functions in o(R) are very
handy in many applications in analysis. Some examples are:
(a) Suppose a < b, and f is the function such that f(x) = 0 if x a or
x b and
f(x) = e
1/(xa)
e
1/(bx)
if a < x < b.
Show that f is indenitely dierentiable on R.
(b) Prove that there exists an indenitely dierentiable function F on R
such that F(x) = 0 if x a, F(x) = 1 if x b, and F is strictly increasing
on [a, b].
(c) Let > 0 be so small that a + < b . Show that there exists an
indenitely dierentiable function g such that g is 0 if x a or x b, g
is 1 on [a +, b ], and g is strictly monotonic on [a, a +] and [b , b].
Solution. Note that the graph of 1/(x a) 1/(b x) is symmetric
about the line x = (a + b)/2 with maximum value of 4/(b a) at the
point x = (a +b)/2.
For (b) consider F(x) = c
_
x

f(t) dt where c is the reciprocal of


_

f(t) dt.
5. (Exercise 5) Suppose f is continuous and of moderate decrease.
(a) Prove that

f is continuous and

f() 0 as [[ .
(b) Show that if

f() = 0 for all , then f is identically 0.
Solution.
53
[

f( +h)

f()[ =

f(x)e
2ix
(e
2ihx
1) dx

[f(x)[[e
2ihx
1[ dx
=
_

[f(x)[2[ sin 2hx[ dx


2A
_

[ sin 2hx[
1 +x
2
dx
Let > 0. Choose N > 0 such that 2A
_
|x|>N
1
1+x
2
dx < /2. Since
| sin 2hx|
1+x
2
0 uniformly on [N, N] as h 0, there exists > 0 such that
2A
_
N
N
| sin 2hx|
1+x
2
dx < /2 for [h[ < . This proves that

f is continuous.
Note: Lebesgue dominated convergence theorem could be applied to yield
a simpler proof.
(b).

f() =
1
2
__

f(x)e
2ix

f(x)e
2i(x+1/2)
dx
_
=
1
2
_

(f(x) f(x
1
2
))e
2ix
dx
Now the result follows from applying Lebesgue dominated convergence
theorem; without it, the proof is harder.
(b). If f, g are of moderate decrease, then the function f(x)g(y)e
2ixy
is
integrable over R
2
. Then it follows from Fubinis theorem that
_
f(x) g(x) dx =
_
g(x)

f(x) dx.
Suppose

f(x) = 0 for all x. For any t and any > 0, K

(tx) as a function
of x is in o(R), so by Corollary 1.10 it is equal to the Fourier transform g

for some g

in o(R). So
_
f(x) g

(x) dx =
_
f(x)K

(t x) dx = 0. Letting
0, we get f(t) = 0.
6. (Exercise 8) Prove that if f is continuous, of moderate decrease, and
_

f(y)e
y
2
e
2xy
dy = 0 for all x R, then f = 0.
Proof.
Let g(x) = e
x
2
. Then (fg)(x) =
_

f(y)e
(xy)
2
dy = e
x
2
_

f(y)e
y
2
e
2xy
dy =
0 for all x. This implies that

f g() =

f() g() =

f()

2
= 0 for
all . So

f = 0. By Theorem 1.9, f = 0.
7. (Exercise 12) Show that the function dened
u(x, t) =
x
t
1
t
(x)
54
satises the heat equation for t > 0 and lim
t0
u(x, t) = 0 for every x,
but u is not continuous at the origin. Proof.
Maple shows that
u
t
=

2
u
x
2
=
x(x
2
6t)
8

t
7/2
e
x
2
/4t
.
It is clear that lim
t0
+ u(x, t) = 0 for all x. If we approach the origin
along the parabola x
2
= t, we get
lim
x0
1
x
2

4
e
1/4
= .
8. (Add in) Suppose f is an even function such that both f and

f are of
moderate decrease. Then

f is also even and the Fourier transform of

f(x)
is f().
Proof.
The
_
below denotes
_

. We have

f() =
_
f(x)e
2ix
dx =
_
f(x)e
2ix
dx =
_
f(x)e
2ix
dx =

f()
so

f is even.
Since
f(x) =
_

f(t)e
2ixt
dt
we have
_

f(t)e
2it
dt =
_

f(t)e
2it
dt =
_

f(t)e
2it
dt = f()
proving that the Fourier transform of

f(x) is f().
9. (Exercise 15) This exercise provides another example of periodization.
(a) Apply the Poisson summation formula to the function g(x) and its
Fourier transform in Exercise 2 to obtain

n=
1
(n +)
2
=

2
(sin )
2
whenever is real, but not equal to an integer.
(b) Prove as a consequence that

n=
1
n +
=

tan
(1)
whenever is real but not equal to an integer.
Solution.
55
By the last item, the Fourier transform of h(x) = g(x) is g(x) in Exercise
2. By Poisson summation formula

n=
sin
2
((n +))

2
(n +)
2
=

n=
g(n)e
2in
= 1.
(If is an integer, then the equality is reduced to trivial 1 = 1.) Since
sin(n +) = (1)
n
sin(), the result follows.
(b). Assume that 0 < < 1. We have from (a), for n ,= 0,
_

0
1
(n +x)
2
+
1
(n +x)
2
dx =
_
1
n +
+
1
n +
_
and since lim
x0

2
(sin x)
2

1
x
2
=
2
/3, it has a removable discontinuity at
x = 0 and lim
x0


tan x
+
1
x
= 0, we have
_

0

2
(sin x)
2

1
x
2
dx =

tan
+
1

.
Therefore
1

n=1
_
1
n +
+
1
n +
_
=

tan
.
Evaluating at = 1/2 yields 0; this is correct since

n=1
4
4n
2
1
=

n=1
2
2n 1

2
2n + 1
is a telescoping series whose sum is 2.
Now we have to prove for any , a noninteger. Let n
1
= |. Then
= n
1
+
1
for some 0 <
1
< 1. Then

n=
1
n +
=

n=
1
n +
1
=

tan
1
=

tan
.
10. (Exercise 19) The following is a variant of the calculation of (2m) =

n=1
1/n
2m
found in Problem 4, Chapter 3.
(a) Apply the Poisson summation formula to f(x) = t/((x
2
+ t
2
)) and

f() = e
2t||
where t > 0 in order to get
1

n=
t
t
2
+n
2
=

n=
e
2t|n|
.
(b) Prove the following identity:
1

n=
t
t
2
+n
2
=
1
t
+
2

m=1
(1)
m+1
(2m)t
2m1
, 0 < t < 1
56
as well as

n=
e
2t|n|
=
2
1 e
2t
1, 0 < t < 1.
(c) Use the fact that
z
e
z
1
= 1
z
2
+

m=1
B
2m
(2m)!
z
2m
,
where B
k
are the Bernoulli numbers to deduce from the above formula,
2(2m) = (1)
m+1
(2)
2m
(2m)!
B
2m
.
Solution.
(a) By Lemma 2.4, the Fourier transform of f(x) = t/((x
2
+t
2
)) is

f() =
e
2t||
. Then by Poisson summation formula with x = 0 ((x+n)
2
= n
2
),
we get (a).
(b).
1

n=
t
t
2
+n
2
=
1
t
+
1

n=0
t
t
2
+n
2
=
1
t
+
2

n=1
t
t
2
+n
2
=
1
t
+
2

n=1
t/n
2
1 + (t/n)
2
=
1
t
+
2

n=1
t
n
2

m=0
(1)
m
t
2m
n
2m
=
1
t
+
2

n=1

m=0
(1)
m
t
2m+1
n
2(m+1)
=
1
t
+
2

n=1

m=1
(1)
m+1
t
2m1
n
2m
=
1
t
+
2

m=1

n=1
(1)
m+1
t
2m1
n
2m
=
1
t
+
2

m=1
(1)
m+1
(2m)t
2m1
57
Next,

n=
e
2t|n|
= 1 + 2

n=1
e
2tn
= 1 + 2
e
2t
1 e
2t
=
1 +e
2t
1 e
2t
=
2 (1 e
2t
)
1 e
2t
=
2
1 e
2t
1
Dividing by 2 on both sides, we get
1
2t
+
1

m=1
(1)
m+1
(2m)t
2m1
=
1
1 e
2t

1
2
.
Letting z = 2t and multiplying both sides by z, we get
1 +
1

m=1
(1)
m+1
(2m)
2z
2m
(2)
2m
=
z
e
z
1
+
z
2
.
Comparing with
z
e
z
1
= 1
z
2
+

m=1
B
2m
(2m)!
z
2m
,
we get
2(2m) = (1)
m+1
(2)
2m
(2m)!
B
2m
.
11. (Exercise 23) The Heisenberg uncertainty principle can be formulated in
terms of the operator L =
d
2
dx
2
+ x
2
, which acts on Schwartz functions
by the formula
L(f) =
d
2
f
dx
2
+x
2
f.
This operator, sometimes called the Hermite operator, is the quantum
analogue of the harmonic oscillator. Consider the usual inner product on
o given by
(f, g) =
_

f(x)g(x) dx whenever f, g o.
(a) Prove that the Heisenberg uncertainty principle implies
(Lf, f) (f, f) for all f o.
58
This is usually denoted by L I.
(b) Consider the operators A and A

dened on o by
A(f) =
df
dx
+xf and A

(f) =
df
dx
+xf.
The operators A and A

are sometimes called the annihilation and cre-


ation operators, respectively. Prove that for all f, g o we have
(i) (Af, g) = (f, A

g),
(ii) (Af, Af) = (A

Af, f) 0,
(iii) A

A = L I.
In particular, this again shows that L I.
(c) Now for t R, let
A
t
(f) =
df
dx
+txf and A

t
(f) =
df
dx
+txf.
Use the fact that (A

t
A
t
f, f) 0 to give another proof of the Heisenberg
uncertainty principle which says that whenever
_

[f(x)[
2
= 1 then
__

x
2
[f(x)[
2
_
_
_

df
dx

2
dx
_

1
4
.
Solution.
(a) By dividing f by
_
(f, f) if necessary, we may assume that (f, f) = 1.
We need to prove that (Lf, f) 1. Now
(Lf, f) = (f

, f) + (x
2
f, f) = (f

, f

) + (x
2
f, f) =
_
[f

[
2
+
_
x
2
[f[
2
= 4
2
_

2
[

f[
2
+
_
x
2
[f[
2
4(
_

2
[

f[
2
)
1/2
)(
_
x
2
[f[
2
)
1/2
4
1
4
= 1
where we have used integration by parts and theorem 4.1 (Heisenberg
uncertainty principle).
(b) Direct checking. Use integration by parts. Note: AA

= L +I.
(c) Similar to (b), we also have (A

t
A
t
f, f) 0 for all t. When this is
written out, we get t
2
(x
2
f, f)t(f, f)(f

, f) = t
2
(x
2
f, f)t+(f

, f

) 0
for all t. It follows that
1 4(f

, f

)(x
2
f, f) 0
from which the result follows.
12. (Problem 1) The equation
x
2

2
u
x
2
+ax
u
x
=
u
t
with u(x, 0) = f(x) for 0 < x < and t > 0 is a variant of the heat
equation which occurs in a number of applications. To solve it, make the
59
change of variables x = e
y
so that < y < . Set U(y, t) = u(e
y
, t)
and F(y) = f(e
y
). Then the problem reduces to the equation

2
U
y
2
+ (1 a)
U
y
=
U
t
,
with U(y, 0) = F(y). This can be solved like the usual heat equation (the
case a = 1) by taking the Fourier transform in the y variable. One must
then compute the integral
_

e
(4
2

2
+(1a)2i)t
e
2i
d.
Show that the solution of the original problem is then given by
u(x, t) =
1
(4t)
1/2
_

0
e
(log(/x)+(1a)t)
2
/(4t)
f()
d

.
Solution.
I have checked that what it says up to the new dierential equation is
correct.
Taking the Fourier transform w.r.t. y of the new equation, we get


U
t
= (4
2

2
+ (1 a)2i)

U.
Solve, using

U(, 0) =

F(), to get

U(, t) =

F()e
(4
2

2
+(1a)2i)t
.
It remains to nd the inverse Fourier transform of e
(4
2

2
+(1a)2i)t
and
then write U(y, t) as a convolution. We have
_

e
(4
2

2
+(1a)2i)t
e
2iy
d =
_

e
4
2

2
t
e
((1a)t+y)2i
d
which is equal to the inverse transform of e
4
2
t
2
evaluated at (1a)t+y.
Using T(e
y
2
) = e

2
and T(f(y)) =

f(
1
) with = 1/

4t, we
nd that the required inverse transform is
1

4t
e
y
2
/4t
evaluating at (1 a)t +y yields
1

4t
e
(y+(1a)t)
2
/(4t)
Therefore
U(y, t) =
_

F()
1

4t
e
(y+(1a)t)
2
/(4t)
d.
Make a substitution = log , and substitute y = log x, we get the
result.
60
13. (Problem 2) The Black-Scholes equation from nance theory is
V
t
+rs
V
s
+

2
s
2
2

2
V
s
2
rV = 0, 0 < t < T, (2)
subject to the nal boundary condition V (s, T) = F(s). An appropri-
ate change of variables reduces this to the equation in Problem 1. Al-
ternatively, the substitution V (s, t) = e
ax+b
U(x, ) where x = log s, =

2
2
(T t), a =
1
2

r

2
, and b =
_
1
2
+
r

2
_
2
reduces (2) to the one-
dimensional heat equation with the initial condition U(x, 0) = e
ax
F(e
x
).
Thus a solution to the Black-Scholes equation is
V (s, t) =
e
r(Tt)
_
2
2
(T t)
_

0
e

(log(s/s

)+(r
2
/2)(Tt))
2
2
2
(Tt)
F(s

)
ds

.
Solution. Multiplying the equation by e
rt
. Dene V
1
= e
rt
V . Then
the equation becomes
V
1
t
+rs
V
1
s
+

2
s
2
2

2
V
1
s
2
= 0.
Let t
1
= (
2
/2)t. Then
V1
t
=
V1
t1
(
2
/2), and upon dividing by
2
/2
on both sides we get
V
1
t
1
=
2r

2
s
V
1
s
+s
2

2
V
1
s
2
.
This is the equation in Problem 1.
For the alternative way, compute
V
t
,
V
s
and

2
V
s
2
, replacing 1/s by e
x
and canceling out e
ax+b
, we get

2
2
(bU+
U

)+r(aU+
U
x
)+

2
2
(a(a1)U+(2a1)
U
x
+

2
U
x
2
)rU = 0.
Since r +

2
2
(2a 1) = 0,

2
2
b +ra +

2
2
a(a 1) r = 0, the coecients
of U and
U
x
are zero. Upon canceling out

2
2
, we get the heat equation
U

=

2
U
x
2
. Since F(e
x
) = F(s) = V (s, T) = e
ax
U(x, 0), the initial
condition for the heat equation is U(x, 0) = e
ax
F(e
x
). From the formula
for solution of heat equation, we get
U(x, ) =
1

4
_

e
ay
F(e
y
)e
(xy)
2
/(4)
dy.
Making a change of variable y = log s

and substitute x = log s, we nd


U(log s, ) =
1

4
_

0
e
a log s

F(s

)e
(log(s/s

)
2
/(4)
ds

.
61
and
V (s, t) =
1

4
_

0
e
a log s+b
e
a log s

F(s

)e
(log(s/s

)
2
/(4)
ds

.
Adding up the exponents of e, we get
a log(s/s

) +b (log(s/s

)
2
/(4)
=
1
4
[log
2
(s/s

) 4a log(s/s

) 4b
2
]
=
1
4
(log(s/s

) 2a)
2
+ (a
2
+b)
Since 4 = 2
2
(T t), 2a = (r

2
2
)(T t) and (a
2
+b) = r(T t),
we are done.
14. (Problem 4) If g is a smooth function on R, dene the formal power series
u(x, t) =

n=0
g
(n)
(t)
x
2n
(2n)!
(3)
(a) Check formally that u solves the heat equation.
(b) For a > 0, consider the function dened by
g(t) =
_
e
t
a
if t > 0
0 if t 0.
One can show that there exists 0 < < 1 depending on a so that
[g
(k)
(t)[
k!
(t)
k
e

1
2
t
a
for t > 0.
(c) As a result, for each x and t the series (3) converges; u solves the heat
equation; u vanishes for t = 0; and u satises the estimate [u(x, t)[
Ce
c|x|
2a/(a1)
for some constants C, c > 0.
(d) Conclude that for every > 0 there exists a nonzero solution to the
heat equation which is continuous for x R and t 0, which satises
u(x, 0) = 0 and [u(x, t)[ Ce
c|x|
2+
.
15. (Problem 7) The Hermite functions h
k
(x) are dened by the generating
identity

k=0
h
k
(x)
t
k
k!
= e
(x
2
/22tx+t
2
)
. (4)
(a) Show that an alternate denition of the Hermite functions is given by
the formula
h
k
(x) = (1)
k
e
x
2
/2
_
d
dx
_
k
e
x
2
. (5)
62
Conclude from the above expression that each h
k
(x) is of the formP
k
(x)e
x
2
/2
,
where P
k
is a polynomial of degree k. In particular, the Hermite functions
belong to the Schwartz space and h
0
(x) = e
x
2
/2
, h
1
(x) = 2xe
x
2
/2
.
(b) Prove that the family h
k

k=0
is complete in the sense that if f is a
Schwartz function, and
(f, h
k
) =
_

f(x)h
k
(x) dx = 0 for all k 0,
then f = 0.
(c) Dene h

k
(x) = h
k
((2)
1/2
x). Then

k
() = (i)
k
h

k
().
Therefore, each h

k
is an eigenfunction for the Fourier transform.
(d) Show that h
k
is an eigenfunction for the operator dened in Exercise
23, and in fact, prove that
Lh
k
= (2k + 1)h
k
.
In particular, we conclude that the functions h
k
are mutually orthogonal
for the L
2
inner product on the Schwartz space.
(e) Finally, show that
_

h
k
(x)
2
dx =
1/2
2
k
k!.
Solution.
(a) Note that e
(x
2
/22tx+t
2
)
= e
x
2
/2
e
(tx)
2
. Note that
_
d
dx
_
k
e
x
2
= p
k
(x)e
x
2
for some degree k polynomial p
k
(x) which is odd if k is odd, and even if k
is even. Therefore the k partial derivative of e
(tx)
2
w.r.t. t, evaluated
at t = 0 is p
k
(x)e
x
2
= (1)
k
p
k
(x)e
x
2
. The formula then follows from
Taylor expansion.
(b) Under the hypothesis, we have by equation (4)
_

f(x)e
(x
2
/22tx+t
2
)
dx = e
t
2
_

f(x)e
x
2
/2+2tx
dx
for all t. By Exercise 8, f = 0.
(c) From equation (4), we get

k=0
h
k
(

2x)
t
k
k!
= exp((x
2
2t

2x +t
2
))
= exp(t
2
) exp((x
_
2/t)
2
)
63
Taking Fourier transform of both sides, we get

k=0

k
()
t
k
k!
= exp(t
2
) exp(2

2it) exp(
2
)
= exp((
2
2(it)

2 + (it)
2
))
=

k=0
h

k
()
(it)
k
k!
Comparing both sides, we get the result.
(d) From the RHS of (4), we have

k=0
L(h
k
(x))
t
k
k!
= L(e
(x
2
/22tx+t
2
)
)
=
d
2
dx
2
e
(x
2
/22tx+t
2
)
+x
2
e
(x
2
/22tx+t
2
)
=
d
dx
e
(x
2
/22tx+t
2
)
(x + 2t) +x
2
e
(x
2
/22tx+t
2
)
= e
(x
2
/22tx+t
2
)
((x + 2t)
2
1) +x
2
e
(x
2
/22tx+t
2
)
= e
(x
2
/22tx+t
2
)
(1 4t
2
+ 4tx)
= e
(x
2
/22tx+t
2
)
+ (4t
2
+ 4tx)e
(x
2
/22tx+t
2
)
=

k=0
h
k
(x)
t
k
k!
+ 2t
d
dt
e
(x
2
/22tx+t
2
)
=

k=0
h
k
(x)
t
k
k!
+

k=0
2kh
k
(x)
t
k
k!
=

k=0
(2k + 1)h
k
(x)
t
k
k!
from which we get Lh
k
= (2k + 1)h
k
. By Exercise 23, L = I +A

A, so it
is Hermitian. Therefore the functions h
k
are mutually orthogonal.
(e) Multiply both sides of equation (4) by h
k
(x) and integrate. On the
LHS, using the orthogonality property, we get
t
k
k!
_

[h
k
(x)]
2
dx.
On the RHS, using equation (5) and integration by parts k times, we get
(1)
k
(1)
k
_

e
x
2
_
d
dx
_
k
e
2txt
2
= (2t)
k
_

e
(xt)
2
dx = 2
k
t
k

.
Therefore result.
64
Chapter 6 The Fourier Transform on R
d
1. (Exercise 1) Suppose that R is a rotation in the plane R
2
, and let
_
a b
c d
_
denote its matrix with respect to the standard basis vectors e
1
= (1, 0)
and e
2
= (0, 1).
(a) Write the conditions R
t
= R
1
and det(R) = 1 in terms of equations
in a, b, c, d.
(b) Show that there exists R such that a +ib = e
i
.
(c) Conclude that if R is proper, then it can be expressed as z ze
i
,
and if R is improper, then it takes the form z ze
i
, where z = x iy.
Solution.
(a) Write D = det(R). Then D = 1 and the conditions R
t
= R
1
implies
_
a c
b d
_
=
_
d/D b/D
c/D a/D
_
,
i.e. we have D = ad bc = 1, a = dD, b = cD, c = bD, d = aD.
(b) From (a) we have D = ad bc = a
2
D (b
2
D) = (a
2
+ b
2
)D which
implies that a
2
+b
2
= 1. Thus (b) follows.
(c). Suppose D = 1. Then ze
i
= (x +iy)(a ib) = (ax +by) +i(bx +
ay) = (ax +by) +i(cx +dy).
If D = 1, then ze
i
= (x iy)(a + ib) = ax + by + i(bx ay) =
ax +by +i(cx +dy). This proves (c).
2. (Exercise 2) Suppose R : R
3
R
3
is a proper rotation.
(a) Show that p(t) = det(R tI) is a polynomial of degree 3, and prove
that there exists S
2
(where S
2
denotes the unit sphere in R
3
) with
R() = .
(b) If T denotes the plane perpendicular to and passing through the
origin, show that
R : T T,
and that this linear map is a rotation.
Solution.
(a). p(t) is clearly a polynomial of degree 3. (The coecient of t
3
is
1.). p(0) = det(R) > 0. Since lim
t
p(t) = , we see that there
exists > 0 such that p() = 0. So R I is singular, and its kernel is
nontrivial.
Chapter 7 Finite Fourier Analysis
1. (Exercise 1.) Let f be a function on the circle. For each N 1 the discrete
Fourier coecients of f are dened by
a
N
(n) =
1
N
N1

k=0
f(e
2ik/N
)e
2ikn/N
=
1
N
N1

k=0
f(e
2ik/N
)e
n
(k), for n Z.
65
We also let
a(n) =
_
1
0
f(e
2ix
)e
2inx
dx
denote the ordinary Fourier coecients of f.
(a) Show that a
N
(n) = a
N
(n +N).
(b) Prove that if f is continuous, then a
N
(n) a(n) as N .
Proof.
(a) is easy since e
2i
= 1.
(b) Note that a
N
(n) is the Riemann sum of the integral a(n) with parti-
tion: k/N, k = 0, , N, and choice of points: k/N, k = 0, , N 1.
2. (Exercise 2) If f is a C
1
function on the circle, prove that [a
N
(n)[ c/[n[
whenever 0 < [n[ N/2.
Proof. Let l be an integer. Then
a
N
(n)[1 e
2iln/N
]
=
1
N
N1

k=0
f(e
2ik/N
)e
2ikn/N

1
N
N1

k=0
f(e
2ik/N
)e
2i(kl)n/N
=
1
N
N1

k=0
f(e
2ik/N
)e
2ikn/N

1
N
Nl1

k=l
f(e
2i(k+l)/N
)e
2ikn/N
=
1
N
N1

k=0
f(e
2ik/N
)e
2ikn/N

1
N
N1

k=0
f(e
2i(k+l)/N
)e
2ikn/N
=
1
N
N1

k=0
[f(e
2ik/N
) f(e
2i(k+l)/N
)]e
2ikn/N
f C
1
implies f is Lipschitz. Let M be its Lipschitz constant. Then
from the above equations,
[a
N
(n)[[1 e
2iln/N
[ M[1 e
2il/N
[
Let l be the integer closest to
N
2n
(in case of ambiguity, choose either
integer). Then [l N/2n[ 1/2, from which we get

ln
N

1
2

[n[
2N

1
4
.
Thus
[1 e
2il/N
[ [2l/N[ = [2ln/N[/[n[
3
2[n[
.
On the other hand, since

ln
N

1
2

1
4
66
implies

2

2ln
N

3
2
,
we have
[1 e
2iln/N
[

2.
Therefore for 0 < [n[ N/2,
[a
N
(n)[
3M
2

2[n[
.
3. (Exercise 3) By a similar method, show that if f is a C
2
function on the
circle, then
[a
N
(n)[
c
[n[
2
, whenever 0 < [n[ N/2.
As a result, prove the inversion formula for f C
2
,
f(e
2ix
) =

n=
a(n)e
2inx
from its nite version.
Solution.
We have
a
N
(n)[2 e
2iln/N
e
2iln/N
]
=
1
N
N1

k=0
[2f(e
2ik/N
) f(e
2i(k+l)/N
) f(e
2i(kl)/N
)]e
2ikn/N

1
N
N1

k=0
M[e
2i(k+l)/N
e
2ik/N
[
2
= M[e
2il/N
1[
2
Then do as we did in Exercise 2 to choose l to reach the conclusion.
For the second part, let N be odd, then for a xed k,

|n|<N/2
a
N
(n)e
2ikn/N
=
1
N

|n|<N/2
N1

j=0
f(e
2ij/N
)e
2ijn/N
e
2ikn/N
=
1
N
N1

j=0
f(e
2ij/N
)

|n|<N/2
e
2ijn/N
e
2ikn/N
= f(e
2ik/N
)
where the last equality follows because

|n|<N/2
e
2ijn/N
e
2ikn/N
= 0
if j ,= k and equal to N if j = k. Note that I have not used the condition
[a
N
(n)[ c/[n[
2
. (to be continued.....).
67
4. (Exercise 4) Let e be a character on G = Z(N), the additive group of
integers modulo N. Show that there exists a unique 0 N1 so that
e(k) = e

(k) = e
2ik/N
for all k Z(N).
Conversely, every function of this type is a character on Z(N). Deduce
that e

denes an isomorphism from



G to G.
Proof.
Let the N roots of unit be 1, ,
2
, cdots,
N1
, where = e
2i/N
.
Let e be a character in Z(N). We have 1 = e(0) = e(1+1+ +1) = e(1)
N
,
so e(1) =
l
for some 0 l N 1. Then for each n Z(N),
e(n) = e(1 + 1 + + 1) = e(1)
n
=
ln
proving that e = e
l
. e
l
e
m
= e
l+m
l +m.
5. (Exercise 5) Show that all characters on S
1
are given by
e
n
(x) = e
inx
with n Z,
and check that e
n
n denes an isomorphism from

S
1
to Z.
Solution.
Since F(0) ,= 0 and F is continuous, there exists > 0 such that c =
_

0
F(y) dy ,= 0. Then
cF(x) =
_

0
F(y)F(x) dy =
_

0
F(y +x) dy =
_
+x
x
F(y) dy
for all x. Dierentiating both sides,
cF

(x) = F( +x) F(x) = F()F(x) F(x) = (F() 1)F(x).


It follows that F(x) = e
Ax
for some constant A. Since + = 0 in S
1
,
we have 1 = F(0) = F( +) = e
2A
. Therefore A = in for some integer
n.
6. (Exercise 6) Prove that all characters on R take the form
e

(x) = e
2x
woth R,
and that e

denes an isomorphism from



R to R.
Solution.
Using the same argument in Exercise 5, F(x) = e
Ax
for some complex
number A. Since [F(x)[ = 1, A must be pure imaginary. So A = 2i for
some real number . e

= e
+
+.
7. (Exercise 8) Suppose that P(x) =

N
n=1
a
n
e
2inx
.
(a) Show by using the Parseval identities for the circle and Z(N), that
_
1
0
[P(x)[
2
dx =
1
N
N

j=1
[P(j/N)[
2
.
68
(b) Prove the reconstruction formula
P(x) =
N

j=1
P(j/N)K(x (j/N))
where
K(x) =
e
2ix
N
1 e
2iNx
1 e
2ix
=
1
N
(e
2ix
+e
2i2x
+ +e
2iNx
).
Observe that P is completely determined by the values P(j/N) for 1
j N. Note also that K(0) = 1, and K(j/N) = 0 whenever j is not
congruent to 0 modulo N.
Solution.
Considering P(x) as a 1-periodic function, we have by Parseval identity:
_
1
0
[P(x)[
2
dx =

N
n=1
[a
n
[
2
. On the other hand, considering f(j) =
P(j/N) =

N
n=1
a
n
e
2inj/N
as a function on Z(N), Parseval identity
yields

N
n=1
[a
n
[
2
= |f|
2
=
1
N

N
j=1
[f(j)[
2
=
1
N

N
j=1
[P(j/N)[
2
.
8. (Exercise 12) Suppose that G is a nite abelian group and e : G C is
a function that satises e(x + y) = e(x)e(y) for all x, y G. Prove that
either e is identically 0, or e never vanishes. In the second case, show that
for each x, e(x) = e
2ir(x)
for some r = r(x) Q of the form p/q, where
q = [G[.
Proof.
Since e(0) = e(0 + 0) = e(0)e(0), we see that either e(0) = 0 or e(0) = 1.
In the rst case, e(x) = e(x + 0) = e(x)e(0) = 0 for all x G. In the
second case, 1 = e(0) = e(x + (x)) = e(x)e(x), proving that e(x) ,= 0
and e(x) = e(x)
1
.
Let N = [G[. For any x G, Nx = x + x + + x = 0, so 1 = e(0) =
e(x)
N
, proving that e(x) is equal to
k
for some k = 0, , N 1, where
= e
2i/N
.
9. (Exercise 13) In analogy with ordinary Fourier series, one may interpret
nite Fourier expansions using convolutions as follows. Suppose G is a
nite abelian group, 1
G
its unit, and V the vector space of complex-valued
functions on G.
(a) The convolution of two functions f and g in V is dened for each a G
by
(f g)(a) =
1
[G[

bG
f(b)g(a b
1
).
Show that for all e

G one has (

f g)(e) =

f(e) g(e).
(b) Use Theorem 2.5 to show that

G
e(c) = 0 whenever c G and c ,= 1
G
.
69
(c) As a result of (b), show that the Fourier series Sf(a) =

f(e)e(a)
of a function f V take the form
Sf = f D,
where D is dened by
D(c) =

G
e(c) =
_
[G[ if c = 1
G
,
0 otherwise.
Since f D = f, we recover the fact that Sf = f. Loosely speaking, D
corresponds to a Dirac delta function; it has unit mass
1
[G[

cG
D(c) = 1,
and (4) says that this mass is concentrated at the unit element in G. Thus
D has the same interpretation as the limit of a family of good kernels.
(See Section 4, Chapter 2.)
Note. The function D reappears in the next chapter as
1
(n).
Solution.
Let N = [G[.

f g(e)
=
1
N

xG
(f g)(x)e(x)
=
1
N
2

xG

bG
f(b)g(xb
1
)e(x)
=
1
N
2

bG
f(b)

xG
g(xb
1
)e(x)
=
1
N
2

bG
f(b)

yG
g(y)e(by)
=
1
N
2

bG
f(b)e(b)

yG
g(y)e(y)
=

f(e) g(e)
(b). If c ,= 1
G
, then there exists e



G such that e

(c) ,= 1.[This can be


proved in two ways. Suppose e(c) = 1 for all e

G. Let H be the cyclic
subgroup generated by c. Since c ,= 1
G
, we have [H[ > 1 so [G/H[ < [G[.
Each e

G induces a character in G/H, and dierent es induce dierent
characters. This is impossible since there are exactly [G/H[ characters on
G/H.
Another proof is to use the fact that G is isomorphic to a direct product of
cyclic groups G1 G2 Gn where [G1[[[G2[[ [[Gn[. Let for each i
70
between 1 and n g
i
be a generator of G
i
, then c =

i
(a
i
g
i
). Since c ,= 0
not all integers a
i
can be zero, let us say its k. Now the map dened by
g
i
1 for i ,= k and g
k
z, with z = e
2i/N
where N = [G
k
[, denes a
character that is not 1 on c (it is z
a
k
).]
Now back to the proof. We have e

G
e =

G
e

e =

G
e since

G is a group. So e

(c)

G
e(c) =

G
e(c). Since e

(c) ,= 1, we get

G
e(c) = 0.
Sf(a) =

f(e)e(a)
=

G
1
N

xG
f(x)e(x)e(a)
=
1
N

xG
f(x)

G
e(a)e(x)
1
=
1
N

xG
f(x)

G
e(a)e(x
1
)
=
1
N

xG
f(x)

G
e(ax
1
)
= (f D)(a)
71