Functional Analysis

Solutions to Extra Exercises

Exercise 1
|f|
C
0,
(I)
= |f|

+ sup
x,yI, x<y
[f(x) f(y)[
[x y[

f(x
0
) +

x
x
0
f

(t)dt

+ sup
x,yI, x<y

b
a
f

(t)
1
[x,y]
[x y[

dt

[f(x
0
)[ + sup
xI

b
a
1
[x
0
,x]
f

(t)dt

+ sup
x,yI, x<y

y
x
1
[x y[
dt

b
a

(t)

p
dt

1
p
[f(x
0
)[ +

L
p
sup
xI
[x x
0
[

L
p
[f(x
0
)[ + ([b a[

+ 1)

L
p
(1)
Exercise 2
(B(, y))
yY
X

is a family of functionals. For any xed x X we have sup

yY
[B(x, y)[ K
x
|y| and hence the family
(B(, y))
y=1
is bounded. The uniform boundedness principle therefore yields sup
y=1
|B(, y)| K < . Altogether we have
[B(x, y)[ = |y|

x,
y
|y|

K|x| |y| (2)

Exercise 3
(i) The following steps are quite similar to the proof of the open mapping theorem. We can write T(X) =

nN
T(B
n
(0)) and want
to prove that for all n N the sets T(B
n
(0)) have empty interior. Assuming for contradiction we have for some n N, y Y, R
that B

(y) T(B
n
(0)). But due to linearity we also have B

(y) T(B
n
(0)) and since T(B
n
(0)) is convex we can conclude
that there exists r R such that B
r
(0) T(B
n
(0)). Since T is not surjective we can nd (scaling if necessary) y , Y with norm
|y| = r. We can nd x
1
B
r
(0) such that |y Tx
1
|
r
2
which clearly fulls y Tx
1
, T(X). Continuing with x
2
B
r
(0)
such that

y Tx
1
T
x
2
2

r
4
we nd a sequence (x
k
)
kN
such that

y T

n
k=1
x
k
2
k1

r
2
n
. Taking the partial sums s
n
we
have constructed a Cauchy sequence which by completeness converges to sum x X. By continuity we also have y = Tx which is
a contradiction.
(ii)

T(X) =

Y = Y ,=
Exercise 4
(i) V = ker() =
1
(|0) is closed as the pre image of a closed set.
(ii) [(x)[ = [(x v)[ || |x v| for any v V and hence taking the innum on both sides yields [(x)[ || dist(x, V ).
On the other hand by the denition of the norm for all n N there exist x
n
X such that |x
n
| = 1 and [(x
n
)[
n
n+1
||.
Dening v
n
:= x
(x)
(x
n
)
x
n
satises v
n
V and [(x)[ =
|(x)|

n
n+1
|(x)|
|(x
n
)|
=
n
n+1
|x v
n
| ||
n
n+1
|| dist(x, V ).
Hence taking the limit proves the equality.
(iii) Assuming the supremum is not attained one gets for all x X such that |x| = 1: dist(x, V ) =
|(x)|

< 1 where (ii) has been

used.
Assuming for contradiction that there exists x X\V and v V satisfying |x v| = dist(x, V ) one may dene y =
xv
xv
X.
This fulls |y| = 1 and dist(y, V ) = inf
uV
|y u| = inf
uV
xu
xv
=
1
xv
dist(x, V ) = 1 where linearity of V has been used.
Assuming for contradiction that the norm is attained there exists y X such that [(y)[ = || |y| which can be decomposed
as y = x v with x X \ V and v V and therefore fulls |x v| = |y| =
|(y)|

=
|(x)|

= dist(x, V ) where (ii) has been used

and completes the proof.
Exercise 5
We have
[(x)[ =

j=1
x
j
2
j

|x|

j=1
1
2
j
= |x|

(3)
1
On the other hand let x
(N)
n
=

1 n N
0 n > N
and compute

(x
(N)
)

x
(N)

= 1

n=N+1
2
n
= 1 2
N

n=1
2
n
= 1 2
N
1 (4)
which implies || = 1. To show that the norm is not attained assume there exists x c
0
such that |x|

= 1 and [(x)[ = 1 and

observe
1 = [(x)[

j=1
[x
j
[
2
j

j=1
|x|

2
j
= 1 (5)
But this implies [x
j
[ = 1 for all j and hence contradicts x c
0
.
Exercise 6
(i) Let any x 7. Due to Bessel > |x|
2

k=1
[/x, x
k
)[
2
which implies /x, x
k
) 0 = /x, 0) and hence by Riezs representation
x
k
0.
(ii) Since

2
0
sin nt sin mt dt =
nm
we have for the orthonormal system

sin nt

nN
and

1
A
7 = L
2
[0, 2] by (i)

1
A
,
sin nt

A
sin nt dt 0. Same argument applies to cos nt.
Exercise 7
Let (x
n
, Tx
n
) any sequence in the graph of T which converges to some (x, y) since XY is complete. Since JT and J are continuous
lim
n
JTx
n
= JTx = J lim
n
Tx
n
which proves due to the injectivity of J that y = lim
n
Tx
n
= Tx i.e. that the graph
of T is closed. Hence the closed graph theorem proves that T is bounded.
Exercise 8
(i) Since the projection is orthogonal each z 7 can be uniquely decomposed as z = z + z with z V and z V

. In this form
the projection reads P
V
( z + z) = z. This yields /x, P
V
y) = / x + x, y) = / x, y) = / x, y + y) = /P
V
x, y)
(ii) P
V
= P
W
P
V
is clear since Im(P
V
) = V W and P
W
[
W
= id. For the other equality use (i) to get /x, P
W
P
V
y) =
/P
W
P
V
x, y) = /P
V
x, P
W
y) = /x, P
V
P
W
y) for all x, y 7 which implies P
V
P
W
= P
W
P
V
.
(iii) Each z 7 can be uniquely decomposed to z = v + w + r such that v V , w W and r V

. With this the

Projection reads P
V W
z = v +w = P
V
z +P
W
z for all z 7 since w +r V

and v +r W

.
(iv) R is closed since the intersection of closed sets is closed. By (ii) we have P
W
P
V
= (idP
V
)P
W
and P
R
= P
R
P
W
. Therefore
it suces to prove (idP
V
)[
W
= P
R
[
W
= P
V
[
W
. For the decomposition as in (i) we have for w W: (idP
V
)w = w = P
V
w.
Exercise 9
The claim is false. For = [1, 1], k(x, y) 1, f 0 there are two dierent solutions u
1
(x) 0 and u
2
(x) = x
Exercise 10
Since
2
3
+
1
3
= 1 we have (L
3/2
)

= L
3
and therefore (

R
f
n
)
nN
is a family of functionals on L
3/2
. By the assumption for all
g L
3/2
we have sup
nN

R
g(x)f
n
(x)dx

< . Applying the uniform boundedness principle yields sup

nN

R
f
n

= C <
which proves

R
g(x)f
n
(x)dx

C |g|
L
3/2
.
Exercise 11
For x, y A exist n
x
, n
y
N such that x
n
x
> 0, y
n
y
> 0 and x
n
= 0, y
m
= 0 for all n > n
x
, m > n
y
. Thus for all t [0, 1] and
for z = tx + (1 t)y we have z
max{n|z
n
=0}
> 0 i.e. z A. Analogues argumentation proves convexity of B. A, B are disjoint by
denition. Now let any 0 ,= x

(
2
)

which due to Riezs representation theorem has the representation x = (x

1
, x
2
, . . . ) with
at least one x
i
,= 0. For the case that exists n > i such that x
n
= 0 dene a =
r
x
i
e
i
+ e
n
A and b =
r
x
i
e
i
e
n
B fullling
x

(a) = x

(b) = r for any r R. For the case x

n
,= 0 for all n > i x some n > i and dene a =
rsgn(x
n
)
x
i
e
i
+
1
|x
n
|
e
n
A and
b =
r+sgn(x
n
)
x
i
e
i

1
|x
n
|
e
n
B fullling x

(a) = x

(b) = r for any r R. We can deduce that neither A nor B are open since
otherwise this would contradict the Banach separation theorem.
2
Exercise 12
Assume X is strictly normed and let any x, y X such that |x| = |y| = 1. Since |(1/2)(x +y)|
x
2
+
y
2
= 1 and for

x
2
+
y
2

= 1 =

x
2

x
2

the property being strictly normed yields x = y with = 1 the only possible choice. A contradiction.
If K contains the zero-element this follows from the norm axioms. Now suppose 0 , K i.e. by closedness of K we have
inf
vK
|v| =: a > 0 and for x, y as in (iii) dene x =
x
a
, y =
y
a
. Therefore due to convexity

x+ y
2

1 which by the
contraposition of (ii) proves x = y.
Assume there exist a, b linear independent such that |a +b| = |a|+|b|. Dene c =
a
a
, d =
b
b
, K := |tc+(1t)d [ t [0, 1]
which is compact and convex and f : [0, 1] R by t |tc + (1 t)d|. By
f(s + (1 )t) = |(s + (1 )t)c + (1 s t +t)d +d d|
= |(sc + (1 s)d) + (1 )(tc + (1 t)d)| f(s) + (1 )f(t)
(6)
for all s, t, [0, 1] it is proved that f is a convex function which by continuity attains its minimum on a compact set which is
unique due to the assumptions. But since
f(0) = |d| = |c| = f(1) = 1 =
|a| +|b|
|a| +|b|
=
1
|a| +|b|
|a +b| =

|x|
|a| +|b|
c + (1
|x|
|a| +|b|
)d

= f

|x|
|a| +|b|

(7)
there are three distinct t [0, 1] such that f(t) = 1. So the uniqueness of the minimum would require some s [0, 1] such that
f(s) < 1 which contradicts the convexity of f.
Exercise 13
(i) Dene X
M
:= |x X [ [f
n
(x)[ M n N =

nN
|x X [ [f
n
(x)[ M for M N which are closed as the intersection
of closed sets. For any x X the sequence f
n
(x) is convergent and therefore bounded i.e. x X
M
for some M N which proves
X =

MN
X
M
. Baires theorem proves that for some M N the set X
M
has non-empty interior. Due to the closedness of X
M
we have proved the existence of V X
M
X open and non-empty fullling the requested property.
(ii) Dene X
M
=

n,m>M
|x X [ [f
n
(x) f
m
(x)[ which are closed. For any xed x X the sequence f
n
(x) is convergent
and therefore Cauchy i.e. there exists M N such that x X
M
and therefore we again have X =

MN
X
M
. An analogous
argument to (i) nishes the prove.
Exercise 14
Denote by

f
n
= /f, e
n
) =

2
0
f(t)e
n
(t)dt the fourier coecients of f with respect to the fourier basis

e
n
(t) =
1

2
e
int

nZ
. The
assumption 0 =

2
0
f(t)dt =

nZ

f
n

2
0
1

2
e
int
dt =

2

f
0
implies

f
0
= 0. By parseval we have

2
0
[f(t)[
2
dt =

nZ\{0}

f
n

nZ\{0}

n
2

f
n

2
=

nZ

2
=

2
0

(t)

2
dt (8)
Equality only holds for n
2
= 1 i.e. for f satisfying

f
n
= 0 for all [n[ > 1. Since f is real-valued by assumption this is equivalent to
f(t) = a sin(t) +b cos(t).
Exercise 15
|T
n
S
n
x TSx| |T
n
S
n
x T
n
Sx| +|T
n
Sx TSx| M |S
n
x Sx| +|T
n
Sx TSx| 0 (9)
where the uniform bounded principle with M = sup
nN
|T
n
| has been used.
Exercise 16
(i) Linearity is trivial, boundedness follows from |Tf|

= |fg|

|f|

|g|

and for f 1 we have |Tf|

= |g|

which
implies |T| = 1.
(ii) Linearity follows from denition. For boundedness note [Tf[

p
i=0
[
i
[ [f(t
i
)[ |f|

p
i=0
[
i
[. In order to prove
optimality of this bound dene
f(t) :=

|
i
|

i
+

[
i+1[

i+1

|
i
|

tt
i
t
i+1
t
i
t
i
t < t
i+1
for i |0, . . . , p
[
p[

p
t = t
p
= 1
(10)
which is continuous has |f|

= 1 and fulls [Tf[ =

p
i=0
[
i
[. Hence |T| =

p
i=0
[
i
[.
(iii) Linearity is clear again. An upper bound is given by [Tf[

1
0
[f(t)[ dt +

p
i=0
[
i
[ [f(t
i
)[ |f|

1 +

p
i=0
[
i
[

. To see
that this bound is indeed optimal denote
i
1
, . . . ,
i
k
those satisfying
i
> 0. For n such that
1
n
<
1
2
min|t
i
l+1
t
i
l
[1 l < k
dene
f
n
(t) = 1
k

l=1

2
2
n

t t
i
l

1
[t
i
l

1
n
,t
i
l
+
1
n
]
(t) (11)
satisfying
i
f(t
i
) = [
i
[, |f
n
|

= 1 and

1
0
f
n
(t)dt 1
2k
n
(12)
which proves |T| = 1 +

p
i=0
[
i
[.
3
Exercise 17
(i) Dene g : span|V,x K by g(v + x) = dist(x, V ) which clearly fulls f[
V
= 0 and f(x) = dist(x, V ). For boundedness
note that
[g(v +x)[ =
[[ |v +x| dist(x, V )
|v +x|
=
|v +x| dist(x, V )

x (
v

|v +x| dist(x, V )
dist(x, V )
= |v +y| (13)
Since there exists (v
n
)
nN
such that |x v
n
| dist(x, V ) and therefore dist(x, V ) = [g(v
n
) g(x)[ |g| |v
n
x| |g| dist(x, V )
we have |g| = 1. Applying Hahn-Banach gives f X

with |f| = 1 and the requested properties.

(ii) Assume V is not dense, i.e. V ,= X. Choose x X \ V which due to the closedness of V fulls dist(x, V ) > 0 and therefore (i)
proves the existence of a nonzero f X

such that f[
V
= 0 which contradicts the assumptions.
Exercise 18
Dene A :
1
R mapping x

jN
[x
2j
[ which is bounded by [Ax[ |x| and therefore its Kernel U = Ker A is closed. For V
dene B :
1

1
mapping (x
1
, x
2
, . . . ) (0, x
2
x
1
, 0, x
4

x
3
2
, 0, . . . ) is bounded since |Bx| 2 |x| and therefore its kernel
V = Ker B is also closed. Every sequence with nite support is clearly contained in the direct sum, which therefore is also dense
in
1
. But for example

1
n
2

nN
, U V since assuming there is a decomposition x = u +v we would have v
2n1
=
n
(2n)
2
for all
n N i.e. |v|

n=1
1
4n
. But since U V is dense we have proved that it cant be closed.
Exercise 19
(i) Linearity is easily seen. For non-continuity dene f
n
= sin nt satisfying |f
n
| = 1 and |Df
n
| = |ncos nt| = n which proves
that D is unbounded.
(ii) Let any sequence (f
n
, f

n
) converging to some (f, g) in norm. The fundamental theorem of calculus yields
f(x) f(0) = lim
n
(f
n
(x) f
n
(0)) = lim
n

x
0
f

n
(t)dt =

x
0
lim
n
f

n
(t)dt =

x
0
g(t)dt (14)
which proves f

= g and therefore (f, g) is in the graph and the graph is closed. Since C([0, 1]) with the supremum norm is a
Banach space, C
1
([0, 1]) cant be a Banach space since otherwise this would contradict the closed graph theorem.
Exercise 20
Since X

is separable S = |f X

[ |f| = 1 X

is also separable and there exists a dense subset |f

n
[ n N S. Pick x
n
such that [f
n
(x
n
)[
1
2
and |x
n
| = 1. Denote V = |

q
n
x
n
[ q
i
Q and assume V ,= X hence there exists x X such that
dist(x, V ) > 0. Due to a corollary to Hahn-Banach there exists nonzero f S vanishing on V with f(x) = dist(x, V ). Hence by
1
2
[f
n
(x
n
)[ = [f
n
(x
n
) f(x
n
)[ |f
n
f| |x
n
| = |f
n
f| the set |f
n
[ n N is not dense in S which is a contradiction and
therefore proves that V is a countable dense subset of X.
Exercise 21
Let any x P. Since

n=1
[x
n
[
2
< there exist only nitely many x
i
1
, . . . , x
i
k
such that x
i
l
>
1
2
. Dene =
1
2
max|1

x
i
l

[ 1
l k and let any y such that |x y| i.e.

n=1
[x
n
y
n
[
2

2
. Thus for all n N [y
n
[ [x
n
[ + < 1 which proves that P
is open.
Exercise 22
(i) For
0
note that [f(0) g(0)[ |f g|

, for I note

1
0
f(x) g(x)dx

1
0
[f(x) g(x)[ dx |f g|

and for J
1
note

1
0
[[f(x)[ [g(x)[[ dx |f g|

. Let any f C([0, 1]) and any > 0. Since the function x [x[
p
is continuous for any p 1
there exists > 0 such that |f g|

< implies [[f(x)[

p
[g(x)[
p
[ < for all x [0, 1] i.e.

1
0
[[f(x)[
p
[g(x)[
p
[ < . Assuming
J
p
is also uniformly continuous for = 1 there would exist some > 0 such that [J
p
f J
p
g[ < 1 for all f, g with |f g| < .
Dene f
n
(x) = n and g
n
(x) = n +

2
with J
p
f
n
= n
p
, J
p
g
n
= (n +

2
)
p
. The inequality (n +

2
)
p
n
p
+ p

2
n
p1
is proved by
comparing their derivatives with respect to :
p
2
(n+

2
)
p1

p
2
n
p1
and the initial equality for = 0. Therefore it is proved that
J
p
g
n
J
p
f
n
= p

2
n
p1
diverges for any and p > 1 i.e. J
p
is not uniformly continuous.
(ii) The space is closed as the intersection of closed sets (two are closed as the preimage of closed sets and the third is closed since it
is the Banach space of hlder-continuous functions). The functions are uniformly bounded since [f(x)[

x 1 and equicontinuous.
Therefore Ascoli-Arzela proves the compactness of its closure, i.e. the compactness of the set itself. So the continuous function J
p
attains its maximum on K.
Exercise 23
Assume D is continuous. The unit ball B
1
(0) E is trivially uniformly bounded. Let any > 0, dene =

D
and observe that
for all x, y [0, 1] such that [x y[ < and all f B
1
(0) E holds that [f(x) f(y)[ = [f

()(x y)[ < |D| = . This proves

equicontinuity and hence by Ascoli Arzela the unit ball in E is relative compact, which implies that E is nite dimensional.
4
Exercise 24
Since

k=1
a
k
< there exists a
k
< 1. Therefore T
k
has due to Banachs x point Theorem a unique x point x
0
. But since
T
k
(T(x
0
)) = T(T
k
(x
0
)) = T(x
0
) and the x point is unique T(x
0
) = x
0
. If T had a dierent x point this would also be a x
point of T
k
which is a contradiction.
Exercise 25
Assuming c is an system we have /x, y) =

iN
x(i)e
i
,

jN
y(j)e
j

iN
x(i) y(i). On the contrary assume that /x, y) =

iN
x(i) y(i) holds for all x, y 7 and that there exists 0 ,= z 7 such that /e
i
, z) = 0 for all i N. But then 0 ,= |z|
2
=
/z, z) =

iN
[/e
i
, z)[
2
= 0 is a contradiction and hence 7 is complete.
Exercise 26
Let 7
2
(x
n
, Tx
n
) (x, y) a Cauchy i.e. convergent sequence. Since 0 /Tx
n
y, z) = /x
n
, Tz) /y, z) /x, Tz) /y, z) =
/Tx y, z) for all z 7 we have |Tx y| = 0 i.e. Tx = y and hence (x, y) is in the graph which proves the closedness of the
graph. An application of the Closed Graph Theorem yields T B(7, 7).
Exercise 27
G is trivially a linear subspace. Now let any f (
p
)

=
q
vanishing on M and for the case that there exists i N such that
f(e
i
) = 0 evaluate f(e
j
e
i
) = 0 = f(e
j
) = f
j
for any j ,= i which proves f = 0. For the case that f(e
i
) ,= 0 for all i N we nd
f(e
i
e
j
) = 0 for all i, j N i.e. f = (1, 1, . . . ) ,
q
which contradicts f . Therefore 17(ii) proves that N is dense.
Exercise 28
|Sf|
2
=

2
1
2
x
2
f(x
1
)
2
dx =
1
2
2
t
2
f(t)
2

1
t
2

dt =

2
1
2
f(t)
2
dt = |f|
2
(15)
Exercise 29
Assume for contradiction that for n N exists x
n
such that | x
n
|
E
> n

|T x
n
|
F
+[ x
n
[

. Dene x
n
=
x
n
x
n

E
for which
the inequality reads

|Tx
n
|
F
+[x
n
[

<
1
n
. Since the terms on the left are non-negative we know that |Tx
n
|
F
and [x
n
[ are
nullsequences. Dene

T : E/N(T) R(T) by x + N(T) T(x) which is bijective and bounded and the inverse mapping
theorem yields |x
n
+N(T)| = inf
yN(T)
|x
n
y|
E
0. Therefore its possible to choose z
n
N(T) such that |x
n
z
n
|
E
<
2dist(x
n
, N(T)) 0 and by assumption also [x
n
z
n
[ < M |x
n
z
n
|
E
0. Using [x
n
[ 0 we see [z
n
[ < [z
n
x
n
[ +[x
n
[ 0.
By the equivalence of ||
E
and [[ on the nite dimensional subspace N(T) we see that also |z
n
|
E
and therefore |x
n
|
E
are
nullsequences which is a contradiction.
5