Você está na página 1de 68

Chapter 1 Fourier series

1.1 Appearance of Fourier series


The birth of Fourier series can be traced back to the solutions of wave equation
in the work of Bernoulli and the heat equation in the work of Fourier.
Consider an elastic string of nite length l xed at the end points x = 0 and
x = l. At time, say t = 0, it is distorted from the equilibrium position and allowed to
vibrate. The problem is to nd the vibrations of the string at any point x and any
time t > 0. The vibration of the string is governed by the linear partial dierential
equation

2
u
t
2
= c
2

2
u
x
2
, u(x, 0) = f(x),

t
u(x, 0) = g(x). (1.1)
Here c
2
denotes a physical constant, f(x) gives the initial position and g(x) gives the
initial velocity.
In 1747, D. Alembert obtained the solution of (1.1) in the form
u(x, t) =
1
2
[f(x +ct) +f(x ct)] +
1
2c
x+ct
_
xct
g(y)dy.
However, in 1753, Bernoulli had a dierent idea of solving this equation which was
based on the observation that the functions sin ct, sin x, and cos ct, cos x satisfy
the equation for any R. If we choose =
n
l
, n = 1, 2, 3, . . . they also satisfy the
1
boundary conditions. As the equation is linear, he argued that any superposition of
such solutions will also be a solution.
To explain this method further, let us use the method of separation of variables.
Let u(x, t) = F(x)G(t). Substituting for u in (1.1), we get c
2
F

(x)G(t) = F(x)G

(t),
which can be rewritten as
F

(x)
F(x)
=
1
c
2
G

(t)
G(t)
. (1.2)
In (1.2), the left hand side is a function of x and the right hand side is a function of
t. This is possible only if
F

(x)
F(x)
= k,
1
c
2
G

(t)
G(t)
= k,
where k is a constant. Thus we end up with two dierential equations,
F

(x) kF(x) = 0, (1.3)


G

(t) kc
2
G(t) = 0. (1.4)
Since the string is xed at end points x = 0 and x = l, we have the boundary
conditions
u(0, t) = 0; u(l, t) = 0, t > 0, (1.5)
which are satised if we assume
F(0) = F(l) = 0. (1.6)
The vibrations of the string at time t depend upon the initial deection and initial
velocity. Let f denote the initial deection and g denote the initial velocity. Thus
the initial conditions are given by
u(x, 0) = f(x), (1.7)
2
u
t
(x, 0) = g(x). (1.8)
Notice that the constant k which appears in (1.3) and (1.4) is a real number. Hence
it can be positive or negative or zero.
Case 1. If k = 0, then (1.3) and (1.4) become F

(x) = 0; G

(t) = 0. Solving for F,


we get F(x) = Ax+B, where A and B are constants. The boundary conditions force
A = B = 0 and hence F = 0.
Case 2. Let k =
2
be positive. Substituting in (1.3), we get F

(x)
2
F(x) = 0. In
this case, the solution is given by F(x) = Ae
x
+Be
x
, where A and B are constants.
Once again the boundary conditions force F to be zero.
Case 3. Let k =
2
. Then (1.3) becomes F

(x) +
2
F(x) = 0. On solving, we get
F(x) = A cos x + Bsin x, where A and B are constants. Applying F(0) = 0, we
get A = 0.
Thus, F(x) = B sin x. The equation F(l) = 0 leads to Bsin l = 0. Then,
we get either B = 0 or sin l = 0. In order to get a non-trivial solution, we assume that
B ,= 0. Thus, sin l = 0, which leads to l = n, n Z. As sin 0 = 0 and sin (n) =
sin n, it is enough to take n N. Thus we have
n
=
n
l
, n = 1, 2, 3, . . . and
F
n
(x) = B
n
sin
n
x.
Now (1.4) becomes G

n
(t) +
2
n
c
2
G
n
(t) = 0. On solving we get,
G
n
(t) = C
n
cos (c
n
t) +D
n
sin (c
n
t).
Let u
n
(x, t) = B
n
sin
n
x(C
n
cos (c
n
t) + D
n
sin (c
n
t)). It is clear that for each n,
u
n
(x, t) satises the wave equation with the correct boundary conditions.
3
The function u
n
also satises the initial conditions
u
n
(x, 0) = B
n
C
n
sin
n
x,
u
n
t
(x, 0) = cB
n
D
n

n
sin
n
x.
Consequently, the superpositions u =

n=1
u
n
solves the wave equation with the bound-
ary conditions u(0, t) = u(l, t) = 0. However, the initial conditions will be satised
only if
f(x) =

n=1
B
n
C
n
sin
n
x,
and g(x) = c

n=1
B
n
D
n

n
sin
n
x. At this point, it is not clear whether any f or g
can be expanded in terms of sin
n
x as above and hence the solution of the wave
equation obtained by Bernoulli was incomplete. In 1807, Fourier was working on the
initial value problem for the heat equation
u
t
(x, t) =

2
u
x
2
(x, t), u(x, 0) = f(x).
Let us assume the same boundary conditions as above, namely u(0, t) = u(l, t) = 0
which means the temperature at the end points is kept at zero throughout. Proceeding
as in the case of wave equation, we get the elementary solutions
u
n
(x, t) = A
n
(sin
n
x) e

2
n
t
.
The superposition u(x, t) =

n=1
u
n
(x, t), which solves the heat equation, satises the
initial condition provided
f(x) =

n=1
A
n
sin
n
x.
4
Fourier asserted that any f satisfying f(0) = f(l) = 0 can be expanded in terms of
sin
n
x as above.
1.2 The formal Fourier series and some special ker-
nels
In Section 1.1, we mentioned the idea of expanding a function f(x) satisfying
f(0) = f(l) = 0 in terms of sin
n
l
x, n = 1, 2, . . . Note that the functions sin
n
l
x are
periodic with period 2l and hence f also has to be periodic in order to be expanded
as above.
Given f with f(0) = f(l) = 0, we can always consider it as a periodic function
of period 2l dened on the whole of R and hence it is reasonable to expect an expansion
of the above type.
For the sake of simplicity, we take l = and consider functions f that are 2-
periodic. Rather than restricting our attention to functions satisfying f(0) = f() =
0, we consider all 2 periodic functions. In this case, besides sin nx, we also need to
include cos nx in the expansion of the type
f(x) =

n=0
(a
n
cos nx +b
n
sin nx).
Using the formulae, cos nx =
1
2
(e
inx
+e
inx
), sin nx =
1
2i
(e
inx
e
inx
), we can rewrite
the above expansion as
f(x) =

n=
c
n
e
inx
.
5
This is a reasonable expansion as f is 2-periodic and so is each e
inx
. The family of
functions e
inx
have the interesting property, which we call orthogonality,

e
inx
e
imx
dx = 0 or 2,
depending on whether n ,= m or n = m. Therefore, if termwise integration is allowed,
then the expansion
f(x) =

n=
c
n
e
inx
,
leads us to
1
2

f(x) e
inx
dx = c
n
.
This formula, due to Fourier, allows us to calculate c
n
in the expansion. We can now
formally introduce the Fourier series.
Denition 1.2.1. Let f be a Lebesgue integrable function on [, ]. Then the n
th
Fourier coecient of f is dened by

f(n) =
1
2

f(x) e
inx
dx, n Z.
The Fourier series of f is formally dened by f(x)

n=

f(n) e
inx
.
Example 1.2.2. Let f(x) = x, x . Then, the Fourier coecients of f are
obtained as follows: for n ,= 0

f(n) =
1
2

x e
inx
dx
6
=
1
2

x d
_
e
inx
in
_
=
1
2in

x d(e
inx
)
=
1
2in
_
_
_
x e
inx

e
inx
dx
_
_
_
=
1
2in
2(1)
n
0 , n ,= 0.
Thus

f(n) =
(1)
n+1
in
, n ,= 0. But

f(0) =
1
2

x dx = 0. In this case, the Fourier


series of f can be formally written as
f(x)

n=
n=0
(1)
n+1
in
e
inx
= 2

n=1
(1)
n+1
n
sin nx.
Example 1.2.3. Let f(x) = x
2
, x . Then the n
th
Fourier coecient of f
is given by

f(n) =
1
2

x
2
e
inx
dx
=
1
2

x
2
d
_
e
inx
in
_
.
Using integration by parts, we get,

f(n) =
2(1)
n
n
2
, n ,= 0.
7
But

f(0) =
1
2

x
2
dx =

2
3
.
In this case, the formal Fourier series of f can be written as
f(x)

2
3
+

n=
n=0
2(1)
n
n
2
e
inx
=

2
3
+

n=1
4(1)
n
n
2
cos nx.
Figure 1.1: Fourier series of x
2
.
8
Consider the Fourier series
f(x)

n=

f(n) e
inx
=

f(0) +

n=1

f(n) e
inx
+

n=1

f(n) e
inx
=

f(0) +

n=1
_

f(n) +

f(n)

cos nx
+i

n=1
_

f(n)

f(n)

sin nx,
using e
inx
= cos nx i sin nx. If f is an even function, then

f will also be an even
function. In fact,

f(n) =
1
2

f(x)e
inx
dx
=
1
2

f(x)e
inx
dx
=
1
2

f(x)e
(inx)
dx
=

f(n),
by applying a change of variable x = y. In this case, the resulting Fourier series
becomes f(x)

f(0) +2

n=1

f(n) cos nx. On the other hand, if f is an odd function,


then

f will also be an odd function and the Fourier series takes the form
f(x) 2i

n=1

f(n) sin nx.


9
1.2.1 Dirichlet kernel
The formal Fourier series will represent the function f provided the partial
sums of the series converges to f. We take the symmetric partial sum
S
N
f(x) =
N

n=N

f(n)e
inx
,
which can be written in the form
S
N
f(x) =
1
2

f(y)D
N
(x y)dy,
where the function D
N
, called the Dirichlet kernel, is given by
D
N
(x) =
N

n=N
e
inx
.
Indeed, from the denition,

f(n)e
inx
=
1
2

f(y)e
in(yx)
dy,
and hence
N

n=N

f(n)e
inx
=
1
2

f(y)
_
N

n=N
e
in(yx)
_
dy.
The kernel D
N
can be explicitly calculated:
D
N
(x) =
sin (2N + 1)
x
2
sin (
x
2
)
. (1.9)
10
Figure 1.2: Dirichlet kernel.
11
In order to establish (1.9), consider
D
N
(x) =
N

n=N
e
inx
=
1

n=N
e
inx
+
N

n=1
e
inx
+ 1
= 1 +
N

n=0
e
inx
+
N

n=0
e
inx
= 1 +
N

n=0
(e
ix
)
n
+
N

n=0
(e
ix
)
n
.
Writing w = e
ix
, we have
D
N
(x) = 1 +
N

n=0
w
n
+
N

n=0
w
n
= 1 +
1 w
(N+1)
1 w
1
+
1 w
(N+1)
1 w
= 1 +
w w
N
w 1
+
1 w
(N+1)
1 w
=
w
N
w
N+1
1 w
=
w
1/2
(w
N
w
N+1
)
w
1/2
(1 w)
=
w
N1/2
w
N+1/2
w
1/2
w
1/2
=
sin (2N + 1)
x
2
sin (
x
2
)
.
Observe that, from the very denition, the Fourier coecients of D
N
are given by

D
N
(n) =
_
_
_
1 if [n[ N
0 otherwise.
12
1.2.2 Dirichlet problem for the disc and Poisson kernel
Let D denote the unit disc in the complex plane i.e. D = z C : [z[ < 1.
Given a continuous function f on the boundary of D we are interested in nding a
function u on D satisfying the Laplace equation u = 0 with the boundary condition
lim
r1
u(re
i
) = f(). This is called the Dirichlet problem for the unit disc D. The
Laplace equation in polar coordinates takes the form

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

2
u

2
= 0.
Integrating the equation against e
in
d we see that
r
2

2
r
2
u
n
(r) +r

r
u
n
(r) +
1
2

2
u

2
(r, )e
in
d = 0,
where
u
n
(r) =
1
2

u(r, )e
in
d.
Integrating by parts, the third term becomes n
2
u
n
(r) and hence we get
r
2

2
r
2
u
n
(r) +r

r
u
n
(r) n
2
u
n
(r) = 0.
Clearly, u
n
(r) = a
n
r
|n|
satises the above equation and the initial conditions lim
r1
u(r, )
= f() leads to lim
r1
u
n
(r) =

f(n). Hence, a
n
=

f(n) and the solution u(r, ) has the
formal expansion
u(r, ) =

n=

f(n)r
|n|
e
in
.
13
As before, we can write this in a compact form. Let us dene a kernel P
r
(), called
the Poisson kernel, by
P
r
() =

n=
r
|n|
e
in
.
Then u(r, ) takes the form
u(r, ) =
1
2

f()P
r
( )d.
The kernel P
r
() can also be calculated explicitly. Indeed, we have
P
r
() =

n=
r
|n|
e
in
= 1 +

n=1
r
n
e
in
+

n=1
r
n
e
in
= 1 +re
i
(1 re
i
)
1
+re
i
(1 re
i
)
1
= 1 +
2r cos 2r
2
1 2r cos +r
2
=
1 r
2
1 2r cos +r
2
.
It can be easily shown that the Fourier coecients of P
r
are given by

P
r
(n) = r
|n|
.
1.2.3 The heat kernel
Returning to the heat equation, let u(x, t) solve the problem
u(x, t)
t
=

2
u(x, t)
x
2
,
14
u(x, 0) = f(x), x .
For each n, let
u
n
(t) =
1
2

u(x, t) e
inx
dx.
Then we have
d
dt
(u
n
(t)) =
1
2

2
u(x, t)
x
2
e
inx
dx,
which after integration by parts, leads to
d
dt
(u
n
(t)) = n
2
u
n
(t).
The solution of this equation is given by u
n
(t) = a
n
e
tn
2
. The initial condition
u(x, 0) = f(x) gives a
n
=

f(n) and hence,
u(x, t) =

n=

f(n)e
tn
2
e
inx
.
As before, we can rewrite this as
u(x, t) =
1
2

f(y)h
t
(x y)dy,
where h
t
(x), called the heat kernel, is given by
h
t
(x) =

n=
e
tn
2
e
inx
.
Unlike the Poisson or Dirichlet kernel, we do not have a closed form expression for
h
t
(x).
15
1.3 Uniqueness of Fourier series and some conse-
quences
Let T denote the unit circle. A function on T can be treated as a function on
R which is 2-periodic. Let c(T) denote collection of all continuous functions on T.
The following theorem gives the uniqueness of Fourier series.
Theorem 1.3.1. Let f c(T). Suppose that

f(n) = 0 n Z then f = 0.
Proof. For 0 < < , to be xed soon, consider the function p

(x) = (1+cos xcos).


Then we claim that p

(x) 1 on [, ] and p

(x) < 1 on [, ]
c
. Since cos x
is an increasing function on [, 0] and a decreasing function on [0, ] we have for
x [, 0], cos x cos 0. Then p

(x) = 1 + cos x cos 1 on [, 0].


On the other hand, if x [0, ] then cos x cos and hence p

(x) 1 on
[0, ]. However, if < x < then cos x < cos () = cos .
Hence, p

(x) < 1 on [, ]. Similarly, if < x < , then cos x < cos which
leads to p

< 1 on [, ].
Let f = g + ih, where g and h are real valued functions on T. We prove that
f(x) = 0 at every x. Without loss of generality, we take x = 0. Suppose f(0) ,= 0.
Then, again without loss of generality, we can assume that g(0) > 0 and we will arrive
at a contradiction using the continuity of g.
Choose > 0 such that g(x) >
1
2
g(0) on [, ]. Dene p

with this and


take P
n
(x) = p

(x)
n
. By the hypothesis,

f(x)P
n
(x)dx = 0,
16
which gives

g(x)P
n
(x)dx = 0.
On the other hand,
_
[,]
c
g(x)P
n
(x)dx goes to 0 by dominated convergence theorem
since p

(x) < 1. This shows that


lim
n

g(x)P
n
(x)dx = 0.
On the other hand, P
n
(x) 1 on this interval and hence,

g(x)P
n
(x)dx g(0).
This contradiction proves our claim that f(0) = 0.
Corollary 1.3.2. Suppose f L
1
(T) and

f(n) = 0 n Z. Then f(x) = 0 a. e.
Proof. Let f L
1
(T). Dene g(x) :=
x
_

f(t)dt, x , then g is continuous


and g

= f a.e. Further, for all n ,= 0, g(n) =


i
b
f
n
so that g(n) = 0.
Applying Theorem 1.3.1 to the continuous function g(x) g(0), we conclude
that g(x) = g(0) is a constant. Hence, f = 0 a.e.
Theorem 1.3.3. Let f c(T) be such that

f l
1
(Z). Then the Fourier series of f
converges uniformly to f on T.
Proof. As

f l
1
(Z),

n=
[

f(n)[ < . Dene S


N
f(x) =
N

n=N

f(n)e
inx
. We claim
that S
N
(f) f uniformly on T.
17
For N > M, we have, for all x T,
[S
N
f(x) S
M
f(x)[ =

n=N

f(n)e
inx

n=M

f(n)e
inx

N|n|>M

f(n)e
inx

N|n|>M
[

f(n)[.
Thus by applying Cauchys criterion for uniform convergence, one can conclude
that S
N
(f) converges to a continuous function g (uniformly) on T. Further g(n) =

f(n) n Z. In fact, for all n


g(n) =
1
2

g(x)e
inx
dx
=
1
2

m=

f(m)e
imx
_
e
inx
dx
=
1
2

m=

f(m)e
i(mn)x
dx
=

f(n).
Here, changing the order of integral and the sum was possible because the convergence
is uniform. As (f g)
b
(n) = 0 n Z, we conclude that f = g.
Corollary 1.3.4. If f c
2
(T), then the Fourier series of f converges uniformly to
f on T.
18
Proof. By the theorem, it is enough to prove that

f l
1
(Z). We shall show that

f(n) = O
_
1
|n|
2
_
as [n[ . Consider, for n ,= 0,

f(n) =
1
2

f(x)e
inx
dx
=
1
2
_
_
_
f(x)
e
inx
in

+
1
in

(x)e
inx
dx
_
_
_
=
1
2in

(x)e
inx
dx
=
1
2in
_
_
_
f

(x)
e
inx
in

+
1
in

(x)e
inx
dx
_
_
_
.
Thus

f(n) =
1
2n
2

(x)e
inx
dx.
Let M = max
x[,]
[f

(x)[. Thus [

f(n)[
M
n
2
for all n ,= 0. Consequently,

n=
[

f(n)[

f(0)

n=0

f(n)

f(0)

+ 2M

n=1
1
n
2
< ,
which shows that

f l
1
(Z).
19
1.4 Convolution theory
Before dening convolution of two functions, we observe the following fact.
Proposition 1.4.1. If f is a periodic function of period 2 , then

f(x +a)dx =

f(x)dx =
+a
_
+a
f(x)dx.
Proof. Consider

f(x+a)dx. Taking x x+a, we get

f(x+a)dx =
+a
_
+a
f(x)dx.
If a > 0, then
+a
_
+a
f(x)dx =

_
+a
f(x)dx +
+a
_

f(x)dx
=

f(x)dx
+a
_

f(x)dx +
+a
_

f(x)dx
=

f(x)dx
+a
_

f(x + 2)dx +
+a
_

f(x)dx
=

f(x)dx
+a
_

f(y)dy +
+a
_

f(x)dx =

f(x)dx.
Similarly, we can prove the result when a < 0.
Denition 1.4.2. For f, g L
1
(T), the convolution of f and g is dened as
(f g)(x) :=
1
2

f(x y) g(y) dy.


20
Immediately we observe that f g = g f. In fact,
(f g)(x) =
1
2
x+
_
x
f(t)g(x t)dt =
1
2

g(x t)f(t)dt = (g f)(x).


(a)
[1,1]
. (b)
[1,1]

[1,1]
.
Figure 1.3
Theorem 1.4.3. L
1
(T) is a commutative Banach algebra.
Proof. We know that L
1
(T) is a Banach space. Let f, g L
1
(T). Consider
|f g|
1
=
1
2

[f g(x)[ dx
=
1
(2)
2

f(x y) g(y) dy

dx

1
(2)
2

_

_

[f(x y)[ [g(y)[ dy


_
dx.
21
Applying Fubinis theorem we get,
|f g|
1

1
(2)
2

_
_

[f(x y)[dx
_
_
[g(y)[dy
=
1
2

|f|
1
[g(y)[ dy
= |f|
1
|g|
1
< .
Thus f g L
1
(T) and |f g|
1
|f|
1
|g|
1
. We have already proved that f g = gf.
In order to prove associativity, consider, for x R,
(f g) h(x) =
1
2

(f g) (y) h(x y) dy
=
1
2

_
1
2

f(t) g(t +y) dt


_
h(x y) dy
=
1
2

f(t)
_
1
2
x+
_
x
g(t u +x) h(u) du
_
dt
=
1
2

f(t) (g h) (x t) dt
= [f (g h)] (x).
By the linearity of integral, it follows that (f + g) h = (f h) + (g h) and
f g = f g. Thus L
1
(T) is a commutative Banach algebra.
Theorem 1.4.4. Let f, g c(T). Then f g c(T).
22
Proof. Let f, g c(T). Let > 0 be given. Let M = max
yT
[g(y)[. For s, t T, we have
[(f g)(s) (f g)(t)[ =
1
2

f(s y) g(y) dy

f(t y) g(y) dy

1
2

[f(s y) f(t y)[ [g(y)[ dy

M
2

[f(s y) f(t y)[ dy.


Since f is uniformly continuous on T, there exists > 0 such that [s t[ < implies
[f(s) f(t)[ <

M
. Thus [(f g) (s) (f g) (t)[ if [s t[ < , proving that
f g c(T).
In general if f, g L
1
(T), then we can show that there exist sequences of
functions f
n
, g
n
in c(T) such that f
n
g
n
converges to f g in L
1
. However, if
we assume that both f and g are bounded, then the convergence becomes uniform,
which is established in the next theorem.
Theorem 1.4.5. Let f, g L
1
(T). If, in addition, f and g are bounded, then there
exist sequences f
n
and g
n
in c(T) such that f
n
g
n
converges to f g uniformly
on T.
Proof. Let f, g L
1
(T) be bounded. Then there exist f
n
and g
n
in c(T) such
that
|f
n
f|
1
0 as n sup
xT
[f
n
(x)[ k
1
, n
23
and
| g
n
g |
1
0 as n sup
xT
[g
n
(x)[ k
2
, n.
Consider
f g f
n
g
n
= f g f
n
g +f
n
g f
n
g
n
= (f f
n
) g +f
n
(g g
n
).
Let M = sup
yT
[g(y)[. For x T, we have,
(f f
n
) g(x) =
1
2

(f f
n
)(x y) g(y) dy
=
1
2

(f(x y) f
n
(x y)) g(y) dy.
Thus
[(f f
n
) g(x)[
1
2

[f(x y) f
n
(x y)[ [g(y)[ dy

1
2
sup
yT
[g(y)[

[f(x y) f
n
(x y)[ dy

M
2
| f
n
f |
1
0 as n .
Also
[f
n
(g g
n
)(x)[ =
1
2

(g g
n
) (x y) f
n
(y) dy

1
2

[g(x y) g
n
(x y)[ [f
n
(y)[ dy

k
1
2

[g(x y) g
n
(x y)[ dy
24
=
k
1
2
| g
n
g |
1
0 as n .
Thus f
n
g
n
f g 0 uniformly on T as n . In other words, f
n
g
n
f g
uniformly on T as n .
Theorem 1.4.6. Let f, g L
1
(T). Then (f g)
b
(n) =

f(n) g(n), n Z.
Proof.
(f g)
b
(n) =
1
2

(f g) (x) e
inx
dx
=
1
2

_
_
1
2

f(x y) g(y) dy
_
_
e
inx
dx
=
1
2

_
_
1
2

f(x y) e
inx
dx
_
_
g(y) dy
=
1
2

_
_
1
2
e
iny

f(x y) e
in(xy)
dx
_
_
g(y) dy
=
1
2

f(n) g(y) e
iny
dy =

f(n) g(n).
1.5 An approximate identity
In the previous section, we have shown that L
1
(T) is Banach algebra under
convolution. It is therefore natural to ask if L
1
(T) admits an identity. In other words,
the question is whether there exists L
1
(T) such that f = f for all f L
1
. It
is easy to see that this is not the case. Indeed, if such a exists, then by Theorem
25
1.4.6

f(n) (n) =

f(n) for all n which is possible only if (n) = 1 for all n. But for
L
1
(T), (n) 0 as n . (See Corollary 1.7.2.) Hence there is no L
1
(T)
such that f = f for all f.
Thus we look for a sequence e
n
such that fe
n
f as n in appropriate
norm. Such a sequence is called an approximate identity.
Towards this end, rst we shall prove the following theorem.
Theorem 1.5.1. Let k
n
be a sequence of functions dened on T satisfying
1.
1
2

k
n
(x) dx = 1 n.
2. There exists 0 M < such that

[k
n
(x)[ dx M, n 1.
3. For every > 0,
_
|x|
[k
n
(x)[ dx 0 as n .
If f c(T), then k
n
f converges to f uniformly on T.
Proof. As
1
2

k
n
(y) dy = 1,
f k
n
(x) f(x) =
1
2

[f(x y) f(x)] k
n
(y) dy,
which gives
[f k
n
(x) f(x)[
1
2

[f(x y) f(x)[ [k
n
(y)[ dy. (1.10)
26
Let > 0 be given. Then by uniform continuity of f there exists > 0 such that
[f(s) f(t)[ < whenever [s t[ < . By (1.10),
[f k
n
(x) f(x)[
1
2

[f(x y) f(x)[ [k
n
(y)[ dy
+
1
2
_
|y|
[f(x y) f(x)[ [k
n
(y)[ dy
=: I
1
+I
2
.
We have,
I
1
=
1
2

[f(x y) f(x)[ [k
n
(y)[dy

[k
n
(y)[ dy
M
2
.
On the other hand
I
2
=
1
2
_
|y|
[f(x y) f(y)[[k
n
(y)[ dy

|f|

_
|y|
[k
n
(y)[ dy, (1.11)
which goes to zero as n . Here |f|

= sup
xT
[f(x)[. Thus f k
n
converges to f
uniformly on T.
Remark 1.5.2. Since c(T) is dense in L
1
(T) and uniform convergence is stronger
than L
1
convergence, it follows from the above theorem that k
n
gives an approxi-
mate identity for L
1
(T).
27
Remark 1.5.3. We shall show that the Dirichlet kernel does not satisfy the hypoth-
esis of Theorem 1.5.1. Recall
D
N
(x) =
N

n=N
e
inx
=
sin (N +
1
2
)x
sin (
x
2
)
.
We have
1
2

D
N
(x) dx = 1. Since
1
2

e
inx
dx = 0 for all n ,= 0. However,
property (ii) of Theorem 1.5.1 fails. We shall show that

[D
N
(x)[ dx c ln N, N.
Consider

[D
N
(x)[dx =

sin (N +
1
2
)x
sin (
x
2
)

dx
2

[sin (N +
1
2
)x[
[x[
dx,
since

sin
x
2
x
2

1. Setting y = (N +
1
2
) x, we get,

[D
N
(x)[dx 2
(N+
1
2
)
_
(N+
1
2
)

sin y
y

dy
= 4
(N+
1
2
)
_
0

sin y
y

dy
= 4
_

_
N
_
0

sin y
y

dy +
(N+
1
2
)
_
N

sin y
y

dy
_

_
.
28
Since the second term on the right hand side is non-negative, we get,

[D
N
(x)[ dx 4
N
_
0

sin y
y

dy
= 4
N1

k=0
(k+1)
_
k

sin y
y

dy.
For k y (k + 1) , we have
1
[y[

1
(k + 1)
. Hence

[D
N
(x)[dx 4
N1

k=0
1
(k + 1)
(k+1)
_
k
[ sin y[ dy
=
4

N1

k=0
1
(k + 1)

_
0
[ sin(y)[dy,
c

N1

k=0
1
(k + 1)
c

l=1
1
l
c

log N, as N .
1.5.1 Fejer kernel
As an example of approximate identity, we introduce the kernel
n
(x). This
is dened, using the Dirichlet kernel D
k
, by

N
(x) =
1
N
N1

k=0
D
k
(x).
29
Note that
f
N
(x) =
1
N
N1

k=0
f D
k
(x)
=
1
N
N1

k=0
S
n
(f(x)), (1.12)
and these are called the Fejer means. Fejer proved the following result. In fact, he
has proved for f L
p
(T), 1 p < . However, we shall give the proof for p = 1
now and give the remaining separately in the solved problem.
Figure 1.4: Fejer kernel.
30
Theorem 1.5.4. Let f L
1
(T). Then f
N
f in L
1
(T).
Proof. The theorem will follow if we can show that
N
is an approximate identity.
In order to show this result, rst we shall calculate N
N
(x).
Consider
N
N
(x) =
N1

n=0
D
n
(x) =
N1

n=0
_
w
n
w
n+1
1 w
_
=
1
1 w
_
N1

n=0
w
n

N1

n=0
w
n+1
_
=
1
1 w
_
1 w
N
1
1
w
w
1 w
N
1 w
_
=
1
1 w
_
w
(1 w
N
)
w 1
w
(1 w
N
)
1 w
_
=
w
(1 w)
2
_
w
N
1 1 +w
N
_
=
w
(1 w)
2
(w
N
+w
N
2)
=
w
(1 w)
2
(w
N/2
w
N/2
)
2
.
Thus

N
(x) =
1
N
_
sin
_
N
x
2
_
sin (
x
2
)
_
2
.
Note that
N
(x) 0 for every x. Further
1
2

N
(x) dx =
1
2

1
N
N1

n=0
D
n
(x) dx
=
1
2N
N1

n=0

D
n
(x) dx = 1.
31
Fix > 0. Let [x[ . Since sin
2
_
x
2
_
sin
2
_

2
_
= C

> 0 we get
N
(x)
sin
2
(N
x
2
)
N C

.
Consider
_
|x|
[
N
(x)[dx =
1
N
_
|x|
sin
2
_
N
x
2
_
sin
2
_
x
2
_ dx

1
N
1
C

_
|x|
sin
2
_
N
x
2
_
dx.

1
N
2
C

0 as N .
Thus
N
satises the required properties of Theorem 1.5.1. Consequently, we con-
clude that whenever f c(T), then f
N
(x) converges uniformly to f on T.
1.5.2 Poisson kernel
Recall that for 0 r < 1,
P
r
() =
1 r
2
1 2r cos +r
2
=

n=
r
|n|
e
in
.
We shall show that P
r
gives an approximate identity. First, notice that P
r
() 0.
(i)
1
2

P
r
() d =
1
2

n=
r
|n|
e
in
_
d
=
1
2

n=

r
|n|
e
in
d
32
=
1
2

n=
r
|n|

e
in
d
= 1.
Figure 1.5: Poisson kernel.
33
(ii) Fix > 0 and consider [[ and
1
2
r < 1. Then
1 2r cos +r
2
= (1 r)
2
+ 2r(1 cos )
= (1 r)
2
+ 4r sin
2
_

2
_
2 sin
2
_

2
_
= 2c

.
Hence,
P
r
() =
1 r
2
1 2r cos +r
2
,

1 r
2
2c

,
and consequently
_
||
P
r
()d
1 r
2
2c

2 0 as r 1.
Thus P
r
satises the required properties for an approximate identity. Consequently,
we conclude that whenever f c(T), f P
r
() converges uniformly to f on T.
1.6 Summability of Fourier series
Suppose we have a series of complex numbers

k=0
c
k
. The n
th
partial sum of
the series is dened as S
n
:=
n

k=0
c
k
. If S
n
s as n , then we say that

k=0
c
k
converges to s. Let
N
=
1
N
(S
0
+ S
1
+ + S
N1
). The arithmetic mean
N
is
called the N
th
partial Cesaro sum of the series

k=0
c
k
. If
N
converges to a limit as
N , then we say that

k=0
c
k
is Cesaro summable to .
A series of complex numbers

k=0
c
k
is said to be Abel summable to s if for
every 0 r < 1, the series

k=0
c
k
r
k
converges and lim
r1

k=0
c
k
r
k
= s.
34
Remark 1.6.1. Convergence Cesaro summability Abel summability. (See more
details in Solved problems.)
Consider
f P
r
() =
1
2

f(x)
_

n=
r
|n|
e
in(x)
_
dx
=
1
2

n=
f(x)r
|n|
e
in(x)
dx
=
1
2

n=

_
f(x)e
inx
_
r
|n|
e
in
dx
f P
r
() =
1
2

n=

f(n)r
|n|
e
in
. (1.13)
Thus we see that f P
r
are nothing but the Abel means of f. Similarly, f
N
are
nothing but the Cesaro means. Hence we have the following theorems.
Theorem 1.6.2. If f c(T), then the Fourier series of f is uniformly Cesaro
summable to f.
Theorem 1.6.3. If f c(T), then the Fourier series of f is uniformly Abel summable
to f.
1.7 Fourier series of f L
2
(T)
Recall that L
2
(T) is a Hilbert space with the inner product
f, g) =
1
2

f(t)g(t)dt for f, g L
2
(T).
35
Theorem 1.7.1. (Bessels inequality) If f L
2
(T), then

n=
[

f(n)[
2

1
2

[f(t)[
2
dt.
Proof. Consider
f S
N
f, f S
N
f) = f, f) f, S
N
f) S
N
f, f) + S
N
f, S
N
f) (1.14)
Recall that S
N
f(x) =
N

n=N

f(n) e
inx
. If we denote e
n
(x) = e
inx
, x R, then we can
write S
N
f =
N

n=N

f(n)e
n
and also

f(n) = f, e
n
). Thus
f, S
N
f) =
_
f,
N

n=N

f(n)e
n
_
=
N

n=N

f(n)f, e
n
)
=
N

n=N

f(n)

f(n)
=
N

n=N
[

f(n)[
2
.
Further
S
N
f, S
N
f) =
_
N

n=N

f(n)e
n
,
N

m=N

f(m)e
m
_
=
N

n=N
N

m=N

f(n)

f(m)e
n
, e
m
).
36
As e
n
, e
m
) =
n,m
, we can conclude that
S
N
f, S
N
f) =
N

n=N
[

f(n)[
2
.
Hence, (1.14) turns out to be
f S
N
f, f S
N
f) = f, f)
N

n=N
[

f(n)[
2
0,
from which it follows that
N

n=N
[

f(n)[
2
f, f)
=
1
2

[f(t)[
2
dt.
Letting N , we get

n=
[

f(n)[
2

1
2

[f(t)[
2
dt.
This leads us to the following important result.
Corollary 1.7.2. (Riemann-Lebesgue lemma) If f L
1
(T), then

f(n) 0 as
n . Equivalently, if f L
1
(T), then

f(x) cos nx dx 0 and

f(x)
sin nx dx 0 as n .
Proof. The result follows immediately from the theorem for f L
2
(T). Let f
L
1
(T). Since L
2
(T) = L
1
(T) L
2
(T) is dense in L
1
(T), there exists a sequence of
functions f
n
in L
2
(T) such that f
n
f in L
1
-norm. Now lim
m

f
n
(m) = 0 for each
n. Consider
[

f
n
(m)

f(m)[
1
2
2
_
0
[f
n
(x) f(x)[ dx
37
= |f
n
f|
1
0,
as n . This shows that

f
n
converges to

f uniformly on Z.
Hence
lim
m

f(m) = lim
m
lim
n

f
n
(m)
= lim
n
lim
m

f
n
(m) = 0.
Theorem 1.7.3. (Best approximation theorem) Let f L
2
(T). Let S
N
(f)
denote the N
th
partial sum of the Fourier series, and t
N
(x) =
N

n=N
c
n
e
inx
be any
trigonometric polynomial. Then |f S
N
f|
2
|f t
N
|
2
and the equality holds i

f(n) = c
n
.
Proof. Consider
|f t
N
|
2
2
= f t
N
, f t
N
) = f, f) f, t
N
) t
N
, f) +t
N
, t
N
). (1.15)
But
f, t
N
) =
_
f,
N

n=N
c
n
e
n
_
=
N

n=N
c
n
f, e
n
)
=
N

n=N
c
n

f(n).
Also
t
N
, t
N
) =
_
N

n=N
c
n
e
n
,
m

n=m
c
m
e
m
_
38
=
N

n=N
m

m=m
c
n
c
m
e
n
, e
m
)
=
N

n=N
m

n=m
c
n
c
m

n,m
=
N

n=N
[c
n
[
2
.
Hence (1.15) becomes
|f t
N
|
2
2
= f, f)
N

n=N
c
n

f(n)
N

n=N
c
n

f(n) +
N

n=N
[c
n
[
2
= f, f)
N

n=N
[

f(n)[
2
+
N

n=N
[

f(n) c
n
[
2
.
Thus
|f t
N
|
2
2
= |f S
N
f|
2
2
+
N

n=N
[

f(n) c
n
[
2
,
proving our assertion. Recall that a complex valued function f on T is said to be
Lipschitz continuous on T if [f(s) f(t)[ < [s t[ for all s, t T. In the follow-
ing theorem, we show that Lipschitz continuity leads to the uniform convergence of
Fourier series.
Theorem 1.7.4. If f is a Lipschitz continuous function on T, then S
N
(f) converges
to f uniformly on T.
Proof. Consider sin (N +
1
2
)t =
1
2i
_
e
iNt
e
it
2
e
iNt
e
it
2
t
_
. As f is Lipschitz contin-
uous, the function
f(xt)f(x)
sin
t
2
is bounded. Now, applying Riemann-Lebesgue lemma
to
f(xt)f(x)
sin
t
2
e
i
2
t
and
f(xt)f(x)
sin
t
2
e
i
2
t
, we can conclude the result.
39
Theorem 1.7.5. Let f c(T). Suppose f is piecewise continuously dierentiable
function on T. Then the Fourier series of f converges uniformly to f on T.
Proof. It is enough to prove that

f l
1
(Z). Given that f is a piecewise continuously
dierentiable function on T. We know that T can be identied with [, ]. Then
there exists a partition x
1
, x
2
, . . . , x
m1
[, ], such that = x
0
x
1

x
m1
x
m
= and f[
[x
j1
,x
j
]
is continuously dierentiable for each j = 1, 2, , n.
Let g
j
denote the continuous derivative of f on [x
j1
, x
j
]. Choose a periodic
function g on [, ] such that g = g
j
on [x
j1
, x
j
]. Then, clearly, g L
2
(T). Using
Bessels inequality, we have

n=
[ g(n)[
2
|g|
2
2
.
Consider, for n ,= 0
x
j
_
x
j1
f(x)e
inx
dx = f(x)
e
inx
in

x
j
x
j1
+
1
in
x
j
_
x
j1
e
inx
f

(x)dx.
Thus

f(n) =
1
2

f(x)e
inx
dx
=
1
2
m

j=1
x
j
_
x
j1
f(x)e
inx
dx.
Now
[

f(n)[ =

g(n)
in

1
2
_
1
n
2
+[ g(n)[
2
_
.
Thus, we get,

n=
[

f(n)[ = [

f(0)[ +

n=0
[

f(n)[
40
[

f(0)[ +
1
2

n=0
_
1
n
2
+[ g(n)[
2
_
< ,
from which it follows that

f l
1
(Z).
Theorem 1.7.6. For any f L
2
(T), S
N
(f) converges to f in L
2
(T), i.e |S
N
f f|
2
0 as N .
Proof. Let > 0 be given. We know that c(T) is dense in L
2
(T). Therefore, given
f L
2
(T), there exists h c(T) such that
|f h|
2
< . (1.16)
Writing f S
N
f = (f h) +(hS
N
h) +S
N
(hf) and noting that |S
N
(hf)|
2

|f h|
2
, it is enough to show that |h S
N
h| < for all N N
0
for some N
0
.
As h c(T), there exists a trigonometric polynomial P of degree N
0
such that
|h P|
2
< . For N N
0
, S
N
P = P and hence
|h S
N
h|
2
|h P|
2
+|S
N
(P h)|
2
2.
This proves our assertion.
Theorem 1.7.7. (Parsevals Theorem) For any f, g L
2
(T),
1
2

f(x)g(x)dx =

n=

f(n) g(n).
41
In particular,
1
2

[f(x)[
2
dx =

n=
[

f(n)[
2
.
Proof. We have shown in Theorem 1.7.6 that S
N
f f in L
2
. Hence, S
N
f, S
N
g)
f, g) which simply means

n=

f(n) g(n) =
1
2

f(x)g(x)dx.
Corollary 1.7.8. The functions e
inx
: n Z forms an orthonormal basis for L
2
(T).
(See Appendix A.1.3.)
Corollary 1.7.9. Let f L
2
(T). Then the Fourier series of f can be integrated term
by term in (a, b) [, ].
Proof. Let g(t) =
[a,b]
(t). Then g L
2
(T) and hence by Parsevals theorem,
1
2

f(t)g(t)dt =

n=

f(n) g(n).
This means that
1
2
b
_
a
f(t)dt =
1
2

n=

f(n)
_
_
b
_
a
e
int
dt
_
_
,
which is our assertion.
42
1.8 Existence of a continuous function whose Fourier
series diverges
We shall show in this section that, there exists a continuous function whose
Fourier series diverges at 0. In order to prove the theorem, we make use of the uniform
boundedness principle from functional analysis. (See Appendix A.2.5.)
Let X = C[, ], Y = R. Dene A
N
f := S
N
f(0). Then [A
N
f[ = [S
N
f(0)[ =
[D
N
f(0)[ |f|

|D
N
|
1
. In other words, the operator norm of A
N
is given by
|A
N
| |D
N
|
1
. We shall show that |A
N
| = |D
N
|
1
. Take g(x) = sgnD
N
(x). Then
|g|

= 1. Although g is not continuous, it can be approximated by a sequence of


continuous functions g
n
.
Choose g
n
such that |g
n
|

= 1. Then A
N
g
n
|D
N
|
1
1, by using Lebesgue
dominated convergence theorem. Thus, |A
N
| = |D
N
|
1
. But |D
N
|
1
c log N (see
Remark 1.5.3). Hence, |A
N
| as N . Hence, it follows from uniform
boundedness principle that there exists a continuous function f in [, ] such that
[A
N
f[ = [S
N
f(0)[ as N , which means the Fourier series of f diverges
at 0.
1.9 Brief history of Fourier series
The appearance of Fourier series can be traced back to the works of d Alem-
bet(1747), Euler (1748) and D.Bernoulli (1753) in the study of vibrating strings, but
the theory of Fourier series truly began with the profound work of Fourier on heat
conduction at the beginning of the nineteenth century. He studied the problem of
heat ow in a thin wire xed at two end points with length and zero temperature
43
at the ends. He showed that the initial temperature can be expressed as the sum of
an innite series of sines and cosines, now popularly known as Fourier series. Even
though Fourier did not give a convincing proof of convergence of such innite series,
he oered the conjecture that convergence holds for an arbitrary function. Subsequent
work by Dirichlet, Riemann, Lebesgue, and others, throughout the next two hundred
years was needed to show that any arbitrary periodic function can be expressed in
terms of trigonometric series.
1.10 Solved problems
1.10.1. Suppose that f C
k
(T). Show that

f(n) = O
_
1
[n
k
[
_
as n .
Solution.

f(n) =
1
2

f(x)e
inx
dx
=
1
2

f(x)d
_
e
inx
in
_
.
Using Bernoullis formula for successive integration by parts, we get

f(n) =
(1)
k
(in)
k
(f
k
)
b
(n).
Then
[

f(n)[
1
[n[
k
[(f
k
)
b
(n)[

1
[n[
k
|f
(k)
|
1
.
44
1.10.2. If f L
1
(T), then show that

f(n) =
1
4

(f(x) f(x +/n)) e


inx
dx. In
addition, if f satises the Holders condition of order , then

f(n) = O
_
1
[n[

_
.
Solution. We have,

f(n) =
1
2

f(x)e
inx
dx.
Since e
i
= 1, we can write

f(n) =
1
2

f(x)e
inx+i
dx
=
1
2

f(x)e
in(x

n
)
dx.
=
1
2
/n
_
/n
f(y +

n
)e
iny
dy.
Hence

f(n) =
1
2

f(x +

n
)e
inx
dx.
Thus
2

f(n) =
1
2

f(x)e
inx

1
2

f(x +

n
)e
inx
dx,
from which it follows that,

f(n) =
1
4

_
f(x) f(x +

n
)
_
e
inx
dx.
45
Then
[

f(n)[
1
4

_
f(x) f(x +

n
)
_
dx

1
4

2
=
k
[n[

,
where k is a constant.
1.10.3. Using Riemann-Lebesgue lemma, prove that lim
R
R
_
0
sin x
x
dx =

2
.
Solution. Let I = lim
R
R
_
0
sin x
x
dx. Let R = (N +
1
2
), then
I = lim
N
(N+
1
2
)
_
0
sin x
x
dx.
Applying change of variable x = (N +
1
2
)y, we get
I = lim
N

_
0
sin (N +
1
2
)y
y
dy.
From the property of Dirichlet kernel, we get

_
0
sin (N +
1
2
)y
sin (
y
2
)
dy =
46
Hence,
I /2 = lim
N

_
0
_
sin (N +
1
2
)y
y
dy
sin (N +
1
2
)y
2 sin (
y
2
)
_
dy
= lim
N

_
0
_
1
y

1
2 sin (
y
2
)
_
sin (N +
1
2
)y dy.
Since,
1
y

1
2 sin
_
y
2
_ L
2
(T), from Riemann-Lebesgue lemma, it follows that I = /2.
1.10.4. Prove the Binomial identity
n

k=0
_
n
k
_
2
=
_
2n
n
_
.
Solution. Consider
n

k=0
_
n
k
_
e
k
(x) = (1 +e
1
(x))
n
.
By Parsavals inequality
n

k=0
_
n
k
_
2
=
1
2

[f(x)[
2
dx =
1
2

[(1 +e
1
(x))
n
[dx
=
1
2

(1 +e
1
(x))
n
(1 +e
1
(x))
n
dx
=
1
2

(1 +e
1
(x))
n
(1 +e
1
(x))
n
dx
=
1
2

(1 +e
1
(x))(1 +e
1
(x))
n
dx
=
1
2

(2 + 2 cos x)
n
dx
47
=
1
2
2
n

(1 + cos x)
n
dx
=
2
2n
2

cos
2n
_
x
2
_
dx
=
_
2n
k
_
.
1.10.5. By considering the Fourier series of f(x) = [x[, x [, ], nd the sum of
the series

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

n=0
1
n
4
. What is the sum of

n=1
1
n
2
.
Solution. We have

f(n) =
1
2

[x[e
inx
dx
=
1
2
_
_
_

[x[ cos nxdx i

[x[ sin nxdx


_
_
_
=
1
2
_
_
_
2

_
0
x cos nxdx
_
_
_
=
1

_
_
_
x sin nx
n

_
0
sin nx
n
dx
_
_
_
if n ,= 0
=
1


1
n

_
0
sin nxdx
=
1
n
2

cos nx

0
=
1
n
2

(1)
n
1.
48
For n ,= 0, we have

f(n) =
_
_
_
0 if n is even
2
n
2

if n is odd

f(0) =
1
2

[x[dx
=
1

_
0
xdx
=
1

_
x
2
2
_

0
=

2
.
Thus the Fourier Series of f(x) = [x[ on [, ] is given by
[x[ =

f(0) +

n=
n=0

f(n)e
inx
=

2
+

n=1,3..
2
n
2

e
inx
=

2
+ 2(2)

n=1
1
(2n 1)
2
e
i(2n1)x
, x [, ]
[x[ =

2

4

n=1
1
(2n 1)
2
e
i(2n1)x
, x [, ]
On putting x = 0, we get
0 =

2

4

n=1
1
(2n 1)
2
49
Hence, it follows that

n=1
1
(2n 1)
2
=

2
8
.
Now we shall calculate the sum of the series

n=1
1
n
2
. Let s =

n=1
1
n
2
. Then
s =

n=2,4
1
n
2
+

n=1,3
1
n
2
=

mN
1
(2m)
2
+

mN
1
(2m1)
2
=
1
4

m=1
1
m
2
+

2
8
Thus
s =
1
4
s +

2
8
.
This leads to

n=1
1
n
2
=

2
6
Using Parsevals inequality, we have
1
2

x
2
dx =

2
4
+

n=
n=0
[

f(n)[
2
.
Hence,
1

_
0
x
2
dx =

2
4
+

n=1,3,...
4
n
4

2
,
which means that
1



3
3
=

2
4
+
8

n=1,3,...
1
n
4
.
50
Thus it follows that

2
3


2
4
=
8

n=0,1,...
1
(2n + 1)
4
,
from which we arrive at

2
12
=
8

n=0,1,...
1
(2n + 1)
4
.
This implies that

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

4
96
.
Next, we shall calculate the sum

n=1
1
n
4
.
Let
s

n=1
1
n
4
.
Then
s

neven
1
n
4
+

n odd
1
n
4
=

n=1
1
(2n)
4
+

n=0
1
(2n + 1)
4
=
1
16

n=1
1
n
4
+

n=0
1
(2n + 1)
4
.
Hence
s


1
16
s

=

4
96
.
51
This leads to

n=1
1
n
4
=

4
90
.
1.10.6. Compute the Fourier coecients of the function f given by
f(x) =
_
_
_
[x[ [x[ < 1
0 otherwise.
Solution. We have

f(n) =
1
2

f(x) e
inx
dx
=
1
2
1
_
1
[x[ e
inx
dx
=
1
2
_
_
_
0
_
1
x e
inx
dx +
1
_
0
xe
inx
dx
_
_
_
=
1
2
_
_
_
xe
inx
in

0
1

0
_
1
e
inx
in
dx
xe
inx
in

1
0
+
1
_
0
e
inx
in
dx
_
_
_
=
1
2
_
e
in
(in)
+
1
in
_
e
inx
in
_
0
1

e
in
(in)
+
1
in
_
e
inx
in
_
1
0
_
=
1
2
_
1
in
_
e
in
e
in
_
+
1
n
2

e
in
n
2

e
in
n
2
+
1
n
2
_
=
1
2
_
1
in
_
e
in
e
in
_
+
2
n
2

1
n
2
_
e
in
+e
in
_
_
=
1
2
_
2
n
2

2 sin n
n

2 cos n
n
2
_
52
=
1
n
_
1
n
sin n
cos n
n
_
.
1.10.7. Compute the Fourier coecients of 1-periodic function f(t) = 1, 0 t 1.
Solution. We have

f(n) =
1
_
0
e
2int
dt
=
e
2int
2in

1
0
=
i
2n
_
e
2in
1

=
i
2n
[(1)
n
1] , n ,= 0
But

f(0) =
1
_
0
dt = 1.
Hence

f(n) =
_
_
_
i
2n
[(1)
n
1] n ,= 0
1 n = 0.
1.10.8. Let f L
1
(T) such that

f(0) = 0. Dene F(t) =
t
_
0
f() d. Show that F is
continuous, 2-periodic and

F(n) =

f(n)
in
, n ,= 0.
Solution. Let > 0 be given. Then for t > s,
[F(t) F(s)[ =

t
_
0
f() d
s
_
0
f() d

53
=

t
_
s
f() d

t
_
s
[f()[ d.
As f L
1
(T), there exists a > 0 such that
t
_
s
[f() [d whenever [t s[ < .
Thus, F is continuous. Consider
F(t + 2) =
t+2
_
0
f() d
=
2
_
0
f() d +
2+t
_
2
f() d
= 2

f(0) +
t
_
0
f() d
= F(t).
Thus F is 2-periodic. By Radon-Nikodym theorem, F is dierentiable with F

(t) =
f(t) f(0). Thus

F(n) =
2
_
0
F(t)e
int
dt
=
2
_
0
F(t) d
_
e
int
in
_
54
= F(t)
e
int
in

2
0

2
_
0
e
int
in
F

(t)dt
=
1
in
2
_
0
f(t)e
int
dt f(0)
2
_
0
e
int
dt
=
1
in

f(n), n ,= 0.
1.10.9. Calculate
[,0]

[0,]
in L
1
(T).
Solution. Note that

[,0]

[0,]
(t) =
1
2
0
_

[0,]
(t s) ds.
Now, for this integral to be non-zero, s should belong to [, 0] and [ +t, t].
If t 0

[,0]

[0,]
(t) =
1
2
t
_

dt =
1
2
[t +].
If 0 < t , then

[,0]

[0,]
(t) =
1
2
0
_
+t
dt =
1
2
[t +].
Hence

[,0]

[0,]
(t) =
1
2
( [t[)
[,]
(t).
1.10.10. Suppose : R T is a continuous group homomorphism then show that
(t) = e
itx
for some real number x. Use this to derive the form of the continuous
group homomorphism : T T.
55
Solution. If is the trivial homomorphism, then we are done. If not, there exists a
a > 0 such that
c =
a
_
0
(t)dt ,= 0.
Thus
c(x) =
a
_
0
(x +t)dt =
a+x
_
0
(t)dt.
So, is dierentiable and

(x) =
1
c
[(a +x) (x)] =
1
c
[(a) 1](x) = c

,
where c

=
1
c
[(a) 1]. Thus, satises the ordinary dierential equation

= c

.
Hence, (t) = e
c

t
and since [(t)[ = 1 t, c

= 2ix for some x R.


Since T R/Z, any continuous group homomorphism : T T gives rise to
a continuous group homomorphism

: R T given by

= , where is the
canonical quotient map, : R R/Z. Also, ker

contains Z. Thus by the form of


the continuous homomorphism from R T, (t) = e
int
for some integer n.
1.10.11. Let G be a closed proper subgroup of T. Show that G is nite.
Solution. Since G is closed, T/G is an abelian group. Let : T/G T be a
continuous group homomorphism. Then the map

: T T dened as

= ,
: T T/G the canonical quotient map, is a continuous group homomorphism of
T to T whose kernel contains G. But by the previous problem,

(t) = e
int
for some
integer n. Therefore ker(

) is nite and hence G is nite.


1.10.12. If k
n
is a sequence of good kernels, show that for 1 p < , |k
n
f
f|
p
0 as n .
56
Solution. Suppose that the conclusion is true for any g C(T) L
p
(T). Let
f L
p
(T). Since k
n
forms a good kernels,
1
2

_
0
[k
n
(t)[dt = c < . As C(T) is
dense in L
p
(T), given > 0, there exists g C(T) such that |f g|
p
<

3c
. By
assumption, there exists an n
0
N such that for all n n
0
|k
n
g g|
p
<

3
. Thus,
|k
n
f f|
p
|k
n
f k
n
g|
p
+|k
n
g g|
p
+|g f|
p
= |k
n
(f g)|
p
+|k
n
g g|
p
+|g f|
p
c|(f g)|
p
+|k
n
g g|
p
+|g f|
p
<

3
+

3
+

3
Thus it is enough to prove for functions in C(T). Let f C(T). Then
|k
n
f f|
p
=
1
2
_
_
2
_
0
[k
n
f(x) f(x)[
p
dx
_
_
1
p
=
1
2
_
_
2
_
0

1
2
2
_
0
k
n
(t)f(x t)dt f(x)

p
dx
_
_
1
p
=
1
2
_
_
2
_
0

1
2
2
_
0
k
n
(t)f(x t)dt
1
2
2
_
0
k
n
(t)f(x)dt

p
dx
_
_
1
p

1
2
_
_
2
_
0
1
2
2
_
0
[k
n
(t) [f(x t) f(x)][
p
dx
_
_
1
p
dt
=
1
2
2
_
0
[k
n
(t)[|f
t
f|
p
dt
=
1
2
2
_

[k
n
(t)[|f
t
f|
p
dt
57
=
1
2

[k
n
(t)[|f
t
f|
p
dt +
2
_

[k
n
(t)[|f
t
f|
p
dt,
using Minkowskis inequality. Now, given > 0 there exists > 0 such that |f
t

f|
p
< , whenever [t[ < . Now, for this there exists an n
0
N such that for all
n n
0
,
1
2
2
_

[k
n
(t)[dt < . Thus
|k
n
f f|
p

1
2
_
_

[k
n
(t)[dt + 2|f|

2
_

[k
n
(t)[dt
_
_
<
2
_
0
[k
n
(t)[dt + 2|f|

= c.
1.10.13. If f is of bounded variation on T, then show that

f(n) = O
_
1
|n|
_
.
Solution. We have
[

f(n)[ =
1
2

2
_
0
e
int
f(t)dt

=
1
2

2
_
0
f(t)d
_
e
int
in
_

=
1
2

f(t)
e
int
in

2
0

2
_
0
e
int
in
df(t)

=
1
2[n[

2
_
0
e
int
df(t)

V ar(f)
[n[
.
58
1.10.14. If the conjugate of Dirichlet kernel is given by

D
N
(x) =

|n|N
sgn(n) e
inx
,
where
sgn(n) =
_

_
1 n > 0
0 n = 0
1 n < 0
show that

D
N
(x)[dx c ln N.
Solution. We know that

D
N
(x) = i
cos (
x
2
) cos (N +
1
2
)x
sin (
x
2
)
.
Then

D
N
(x)

dx =

cos (
x
2
) cos (N +
1
2
)x
sin (
x
2
)

dx
=

cos (
x
2
) cos Nx cos(
x
2
) + sin Nx sin (
x
2
)
sin (
x
2
)

dx
=

cos (
x
2
)[1 cos (Nx)]
sin (
x
2
)
+ sin Nx

dx

1 cos (Nx)
sin (
x
2
)

dx + 2
= 2

sin
2
_
Nx
2
_
sin (
x
2
)

dx + 2
59
= 4

2
_

sin
2
x
sin x

dx + 2
= 8

2
_
0

sin
2
(Nx)
sin x

dx + 2
8

2
_
0

sin
2
Nx
2x

dx + 2
= 4

2
_
0

sin
2
Nx
x

dx + 2
= 4
N
2
_
0

sin
2
y

[y[
dy + 2
= 4
N
2
_
0
[sin y[
[y[
dy + 2
= 4
_

2
_
0

sin x
x

dx +
N

n=2
n
2
_
(n1)

2
1
x
dx + 2
_

_
= c +
N

n=2
ln x[
n
2
(n1)

2
+ 2
= c +
N

n=2
ln
n

2
(n 1)

2
+ 2
= c +
N

n=2
ln
_
n
n 1
_
= c + ln
_
N

n=2
n
n 1
_
60
= c + ln N
c ln N.
1.10.15. Let f C(). For each pair of integers m, n, where 0 m < n, dene

m,n
(f, x) =
1
nm
n

i=m+1
S
i
(f, x), where S
i
(f, x) stands for i-th partial sum of f at x.
(a) Show that
kn,(k+1)n
(f; x) f(x) uniformly as n , where k, n are integers.
(b) If k, m, n are positive integers with kn m (k+1)n show that [
kn,(k+1)n
(f; x)
S
m
(f; x)[
2A
k
, where f is such that [

f(j)[
A
|j|
for j ,= 0.
(c) Using (a) and (b) show that S
n
(f; x) f(x) uniformly on T whenever

f(n) =
O(
1
n
).
Solution. For integers m and n, we have
m,n
(f; x) =
1
n m
n

i=m+1
S
i
(f; x). Rewrit-
ing this in terms of the kernel, we obtain

m,n
(f; x) =
1
n m
[(n + 1)
n+1
(f; x) (m+ 1)
m+1
(f; x)] (1.17)
Using the Fourier expansion of the kernel, one obtains the following Fourier expansion
for
m,n
(f) as follows:

m,n
(f; x) = S
m
(f; x) +

m<|j|n
n + 1 [j[
n m

f(j)e
ijx
. (1.18)
By (1.17), we have

kn,(k+1)n
(f; x) =
1
n
[(k + 1)n + 1)
(k+1)n
(f; x) (kn + 1)
kn+1
(f; x)]
= (k + 1 +
1
n
)
(k+1)n
(f; x) (k +
1
n
)
kn+1
(f; x) (1.19)
61
which converges uniformly to (k+1)f(x)kf(x) = f(x) as n , by using problem
1.10.12. This proves (a).
To prove (b), we use (1.18). In fact, from (1.18), we get

kn,(k+1)n
(f; x) S
m
(f; x)

|kn||j|(k+1)n
[

f(j)[ 2
(k+1)n

j=kn+1
A
j

2A
k
.
For (c), given that

f(n) = O(
1
n
), that is [

f(n)[
A
[n[
, for n ,= 0. Let > 0 be given.
Choose an integer k > 0 such that
A
k
<

4
. By (a) we can nd n
0
k such that for
all n n
0
, [
kn,(k+1)n
(f; x) f(x)[ <

2
. Let m kn
0
such that for some n n
0
,
kn m < (k + 1)n. By (b), we have [
kn,(k+1)n
(f; x) S
m
(f; x)[ <
2A
k
<

2
. Thus
by the above two inequalities, (c) follows.
1.10.16. Show that

n=1
(1)
n1
is Cesaro summable to
1
2
.
Solution.

n=1
(1)
n1
. Here a
n
=
_
_
_
1 if n is odd
1 if n is even.
Here s
n
is the n-th partial sum of a
n
and s
n
=
_
_
_
1 if n is odd
0 if n is even.
Consider

n
=
s
1
+s
2
+ +s
n
n
.

2n
=
s
1
+s
2
+ +s
2n
2n
=
n
2n
62
=
1
2
n N.
lim
n

2n
=
1
2

2n1
=
s
1
+s
2
+ +s
2n1
2n 1
=
n 1 + 1
2n 1
=
n
2n 1
.
Therefore
2n1

1
2
as n
lim
n

n
=
1
2
.
Hence

n=1
(1)
n1
is Cesaro summable to
1
2
.
1.10.17. Show that

n=0
(1)
n
(n + 1) is not Cesaro summable but Able summable to
1
4
.
Solution. We know that

n=0
(1)
n
(n + 1)r
n
= 1 2r + 3r
2
= (1 + r)
2
for
0 < r < 1. Then
lim
r1

n=0
(1)
n
(n + 1)r
n
= lim
r1
(1 +r)
2
=
1
4
.
Therefore,

n=0
(1)
n
(n + 1) is Abel summable to
1
4
. Now consider the series

n=0
(1)
n
(n + 1) =

n=1
(1)
n1
n.
Here s
n
=
_
_
_
n
2
if n is even
n+1
2
if n is odd.
Clearly,

2n
=
s
1
+s
2
+ +s
2n
2n
= 0 for all n,
63

2n1
=
s
1
+s
2
+ +s
2n1
2n 1
=
2n 1 + 1
2(2n 1)
=
n
2n 1
.
Therefore,
2n1

1
2
as n . Hence

n=0
(1)
n
(n + 1) is not Cesaro summable.
1.10.18. Prove that if

i=1
c
i
converges to s, then

i=1
c
i
is Cesaro summable to s.
Solution. Let s
n
=
n

i=1
c
i
. The assumption is that s
n
s as n . Given > 0
there exists m N such that [s
n
s[ < for all n > m. Now for all n > m, we can
write
[
n
s[ =

s
1
+s
2
+ +s
n
n
s

(s
1
s) + + (s
m
s) + (s
m+1
s) + + (s
n
s)
n

i=1
(s
i
s)
n
+
n

i=m+1
(s
i
s)
n

A
n
+
n m
n

A
n
+,
where A =
m

i=1
(s
i
s). Now we can nd m
1
N such that
A
n
< for all n m
1
.
Choose m
0
= maxm, m
1
. Then [
n
s[ < for all n m
0
. The series

n=1
c
n
is
Cesaro summable to s.
64
1.10.19. Assume that

n=1
a
n
is Cesaro summable to l and lim
n
na
n
= 0. Then show
that

n=1
a
n
converges to l.
Solution. Assume that

n=1
a
n
is Cesaro summable to l and lim
n
na
n
= 0. Consider,
s
n
=
n

i=1
a
i
and
n
=
s
1
+s
2
+ +s
n
n
.
By the assumption, lim
n

n
= l

n
=
s
1
+s
2
+ +s
n
n
=
a
1
+ (a
1
+a
2
) + + (a
1
+a
2
+ +a
n
)
n
=
n

i=1
_
1
i 1
n
_
a
i
.
Now

n
s
n
=
n

i=1
_
1
i 1
n
_
a
i

n

i=1
a
i
=
1
n
n

i=1
a
i

1
n
n

i=1
i a
i
=
s
n
n

1
n
n

i=1
i a
i
_
1 +
1
n
_
s
n
=
n
+
1
n
n

i=1
ia
i
(1.20)
Since lim
n
na
n
= 0, lim
n
1
n
n

i=1
i a
i
= 0. Taking limit as n in 1.20, we have
lim
n
_
1 +
1
n
_
s
n
= lim
n

n
= l. Thus lim
n
s
n
= l.
Here we have used the following result. If lim
n
a
n
= m ,= 0, lim
n
a
n
b
n
= l then
lim
n
b
n
exists and equals
l
m
.
65
1.10.20. If

n=1
C
n
is Cesaro summable to l, then prove that

n=1
C
n
is Abel summable
to l.
Solution. Assume that

n=1
C
n
is Cesaro summable to l. Dene s
n
and
n
as before
then lim
n

n
= l. It can be easily seen that C
1
= s
1
; C
n
= s
n
s
n1
for all n 2 and
s
1
=
1
, s
n
= n
n
(n 1)
n1
for all n 2. Consider 0 < r < 1. Now it can be
easily seen that

n=1
n
n
r
n1
converges absolutely for all 0 r < 1. Then
(1 r)
2

n=1
n
n
r
n1
= (1 r)

n=1
[n
n
r
n1
n
n
r
n
]
= (1 r)
_

n=1
n
n
r
n1

n=1
n
n
r
n
_
= (1 r)
_

1
+

n=2
n
n
r
n1

n=2
(n 1)
n1
r
n1
_
= (1 r)
_
s
1
+

n=2
(n
n
(n 1)
n1
) r
n1
_
= (1 r)
_
s
1
+

n=2
s
n
r
n1
_
= (1 r)

n=1
s
n
r
n1
=

n=1
s
n
r
n1

n=1
s
n
r
n
= s
1
+

n=2
s
n
r
n1

n=2
s
n1
r
n1
= s
1
+

n=2
(s
n
s
n1
) r
n1
= c
1
+

n=2
C
n
r
n1
=

n=1
C
n
r
n1
.
66
Since

n=1
n
n
r
n1
converges for 0 r < 1, it follows that

n=1
C
n
r
n1
converges for
0 r < 1. Dene f(r) =

n=1
C
n
r
n1
for all 0 r < 1. Now we shall show that
lim
r1
f(r) = l. Choose > 0. Then there exists m N such that [
n
l[ < for all
n m. Now choose = . Then for all 1 < r < 1, we have
[f(r) l[ =

n=1
C
n
r
n1
l

(1 r)
2

n=1
n
n
r
n1
l(1 r)
2

n=1
nr
n1

= (1 r)
2

n=1
n(
n
l)r
n1

(1 r)
2

n=1
n[
n
l[ r
n1
= (1 r)
2
_
m1

n=1
nr
n1
[
n
l[ +

n=m
nr
n1
_
= (1 r)
2
_
m1

n=1
n[
n
l[ +(1 r)
2
_
= k(1 r)
2
+, where k =
m1

n=1
n[
n
l[
k
2
+.
Now for any

> 0, there exists unique > 0 (in fact other root is negative) such
that k
2
+ =

. Hence lim
r1
f(r) = l.
1.10.21. Assume that f is periodic and monotone on [, ].Then show that

f(n) =
O
_
1
|n|
_
.
67
Solution. Assume that f is monotone. Then by Lebesgue-Young theorem f is dif-
ferentiable a.e and f

L
1
[, ]. Now

f(n) =
1
2

f(t) e
int
dt
=
1
2
_
_
f(t)
e
int
in

(t)
e
int
i n
dt
_
_
. (1.21)
The rst term of the R.H.S of (1.21) is zero since f() = f(). Then

f(n) =
1
2in

(t)e
int
dt which implies

f(n)


1
2[n[

[f

(t)[ dt
=
K
2[n[
, since f

L
1
[, ]

f(n) = O
_
1
[n[
_
.
1.11 Exercises
1.11.1. Let f L
1
(T). Show that lim
0

[f(x +) f(x)[ dx = 0.
1.11.2. Show that Fourier series of the function
f(x) =
_
_
_
1 x is rational
0 x is irrational
is identically 0 and does not converge to f for any rational number x.
68

Você também pode gostar