< ;

Functional Analysis: Homework 2

HADIOUCHE Azouaou - 3 ING: Group 3

Problem 1
Let H be a Hilbert space, (ui )i∈I an orthonormal system in H with I arbitrary index set.

Question (1). Give an orthonormal system in l2 and L2 ([0, 1], C).

• l2 : consider ei = (δi,n )n∈N where δij is the Kronecker delta, the family {ei }i∈N is an
orthonormal system, and even more, and orthonormal basis. Let n, m ∈ N, we have
 1/2

||en ||2 =  |(en )i |2  =1

X

i≥1

⟨en , em ⟩ = (en )i (em )i = 0

X

i≥0

• L2 ([0, 1], C): consider the family (fn )n∈N defined as

n(n + 1) if n1 > x >

(p
1
fn = n+1
0 otherwise

let n, m ∈ N we have
Z 1 1/2
||fn ||2 = |fn (x)|2 dx
0
Z 1/n !1/2
= n(n + 1) dx
1/(n+1)

1
 1/2
= n(n + 1) =1
n(n + 1)
and also we have
Z 1
⟨fn , fm ⟩ = fn (x)fm (x)dx = 0
0

that is because if n ̸= m then fn (x) ̸= 0 ⇔ fm (x) = 0.

Question (2-3). Showing that for any scalars (ai )i∈I ⊂ K we have that
n 2 n
=
X X
ak uk |ak |2
k=1 k=1

and deduce that an orthonormal family is linearly independent.

< ;8< ;
1
< ;

We have
n 2 n n
=⟨
X X X
ak uk ak uk , ak uk ⟩
k=1 k=1 k=1
n X n
=
X
ak al ⟨uk , ul ⟩
k=1 l=1
n X n
=
X
ak al δlk
k=1 l=1
n
=
X
|ak |2
k=1

now we prove that it is linearly independent, we have

n n n
ak uk = 0 ⇒ ak uk = 0 ⇒ |ak |2 = 0 ⇒ ∀k ∈ N, ak = 0
X X X

k=1 k=1 k=1

thus it is linearly independent.

Question (4). Prove the Bessel inequality stating

X
|⟨x, ui ⟩|2 ≤ ||x||2
i∈I

with

|⟨x, ui ⟩| = sup
X X
|⟨x, uj ⟩|
i∈I J⊂I j∈J
J finite

Let J ⊂ I with J finite, from the previous question we have that || j∈J ⟨x, uj ⟩uj ||
2 =
P

j∈J |⟨x, uj ⟩| , we have the following

P 2

= ⟨x −
X X X
x− ⟨x, uj ⟩uj ⟨x, uj ⟩uj , x − ⟨x, uj ⟩uj ⟩
j∈J j∈J j∈J
2 * +
= ||x|| + − 2 x,
X X
2
⟨x, uj ⟩uj ⟨x, uj ⟩uj
j∈J j∈J

= ||x||2 + |⟨x, uj ⟩|2 − 2

X X
|⟨x, uj ⟩|2
j∈J j∈J

= ||x|| − |⟨x, uj ⟩| > 0

X
2 2

j∈J

thus we obtain that ||x||2 ≥ |⟨x, uj ⟩|2 for any J finite, thus by taking the sup on
P
j∈J
both sides we obtain that
sup |⟨x, ui ⟩|2 = sup
X
|⟨x, uj ⟩| ≤ ||x||2
i∈I J⊂I j∈J
J finite

< ;8< ;
2
< ;

Question (5). Deducing that S : H → l2 , x 7→ (⟨x, ui ⟩)i∈I is linear and

continuous.

The linearity of S comes from the fact that ⟨., .⟩ is a bilinear map, so by fixing the
second parameter, we have a linear map for each of the terms of the sequence, thus a
linear operator. The continuity of S comes directly from the fact that it just needs to be
bounded since it is linear and from the Bessel inequality proved above.

Question (6). Let x ∈ H such that x = n≥1 an un . Show that ∀n ≥

P

1, an = ⟨x, un ⟩.

Let x ∈ H such that x = we have for any n ∈ N

P
n≥1 an un
* +
⟨x, un ⟩ = = ai ⟨ui , un ⟩ = ai δin = an .
X X X
ai ui , un
i≥1 i≥1 i≥1

which concludes the proof.

Question (7). Let I = N, and define Fn = span({ui }ni=1 ), and consider

x = k≥1 ak uk . Show that PFn (x) = nk=1 ak uk .
P P

Since Fn is finite dimension then it is closed, and we have that H = Fn ⊕ Fn⊥ so for any
x ∈ H, x = xFn + xFn⊥ and also that xFn = PFn (x). Since xFn ∈ Fn then xFn = nk=1 bk uk
P

we obtain for k ∈ 1, n

bk = ⟨xFn , uk ⟩ = ⟨xFn , uk ⟩ + ⟨xFn⊥ , uk ⟩ = ⟨x, uk ⟩ = ak

from the fact that ⟨xFn⊥ , uk ⟩ = 0 since it is orthogonal on all vectors of Fn in particular
uk with k ∈ 1, n. Thus we obtain that PFn (x) = nk=1 ak uk .
P

Question (8). Let (an )n∈N ∈ l2 , prove that is convergent in H

P
n≥1 an un
and deduce that S is surjective.

We have that (an )n∈N ∈ l2 then we have that (an )n∈N ∈ l1 so n≥1 |an | converges, since
P

H is Banach, then any absolutely convergent series is convergent, and we have

||an un || =
X X
|an |
n≥1 n≥1

thus the series n≥1 an un is convergent to some x and we have clearly that S(x) = (an )n∈N
P

thus S is surjective.

< ;8< ;
3
< ;

Question (9). Let (un )n∈N a Hilbertian basis of H, prove that ∀x, y ∈ H

x=
X
⟨x, un ⟩un
n≥1

⟨x, y⟩ =
X
⟨x, uk ⟩⟨y, uk ⟩
k≥1

||x||2 =
X
|⟨x, uk ⟩|2
k≥1

Let Sn = nk=1 ⟨x, uk ⟩uk . We have from the question (4) that k∈N |⟨x, uk ⟩|2 ≤ ||x||2
P P

and since the space is Hilbert then it is complete so any absolutely convergent series is
convergent. Thus we obtain that limn→∞ Sn = S. We need to prove that S = x, for that,
we prove that S − x is orthogonal to every element uk and since it is dense in H then
S − x = 0. We have for k ∈ N

⟨S − x, uk ⟩ = ⟨ lim Sn , uk ⟩ − ⟨x, uk ⟩
n→∞
= lim ⟨⟨x, uk ⟩uk , uk ⟩ − ⟨x, uk ⟩
n→∞
= ⟨x, uk ⟩ − ⟨x, uk ⟩ = 0

then we get that S = x. We have

* +
⟨x, y⟩ =
X X
⟨x, uk ⟩uk , ⟨y, uk ⟩uk
k≥1 l≥1

=
XX
⟨x, uk ⟩⟨y, ul ⟩⟨uk , ul ⟩
k≥1 l≥1

=
X
⟨x, uk ⟩⟨y, uk ⟩
k≥1

and also replacing y = x we get

||x||2 = ⟨x, x⟩ =
X
|⟨x, uk ⟩|2
k≥0

Question (10). Deduce that S defines an isomorphism.

We prove that ⟨Sx, Sy⟩ = ⟨x, y⟩, we have

⟨Sx, Sy⟩ = ⟨x, uk ⟩⟨y, yk ⟩ = ⟨x, y⟩

X

k≥1

thus S is an isometry, so it is an open map, by the Open mapping theorem we have that
S is surjective. Thus S is an isomorphism between H and l2 .

< ;8< ;
4
< ;

Question (11). Prove that if a Hilbert space is seperable then it has a

Hilbertian basis.

Let H be seperable, then there exists a Q ⊂ H such that Q is countable and Q = H. Q

has a basis in H and it would be at most countable, by the Gram-Schmidtt algorithm will
give us a countable orthonormal basis B such that B.

Question (12). Deduce that any two seperable Hilbert spaces are isomor-
phic.

From question (11) we have that any two seperable Hilbert spaces have a Hilbertian basis,
and thus from question (10) they are isomorphic to l2 , thus by the transitivity of the
isomorphism relation, the two spaces are isomorphic.

Problem 2
R1
Let E = C([0, 1], C) with the inner product ⟨f, g⟩ = 0 f (t)g(t)dt. And consider the
following spaces
 Z 1 
E0 = f ∈ E | f (t)dt = 0 H = {f ∈ E | f (1) = 0} H0 = H ∩ E0
0

Question (1). Showing that H0 is a closed proper subspace of H.

We have clearly that E0 and H are vector spaces, then their intersection is a vector space
so H0 is a vector subspace of H since H0 ⊂ H. We have that 0 ∈ H0 so H0 ̸= ∅ and
f (t) = 1 − t ∈ H \ H0 thus we obtain that H0 is a proper subspace of H. Consider the
following operator
T0 : H → C
Z 1
f 7→ f (t)dt
0

We have that T0 is continuous in H since T0 (f ) = ⟨f, 1⟩, then H0 = T0−1 ({0}) is closed in
H. So we conclude that H0 is a closed proper subspace of H.

Question (2). Let f1 ∈ E defined as f (t) = t − 1/2.

a) Show that E = span(H, f1 ).

b) Show that E0 = span(H0 , f1 ).

c) Show that f1 ∈ H0 .

a – We have that H ∪ {f1 } ⊂ E thus span(H, f1 ) ⊂ span(E) = E, thus we prove the

other inclusion. Let f ∈ E, we will show that ∃h ∈ H such that f (t) = λf1 (t) + h(t).
< ;8< ;
5
< ;

We have that f (1) = λf1 (t) + h(1) = λ(1 − 1/2) = λ/2 thus we have that λ = 2f (1).
Thus we obtain that f (t) = 2f (1)f1 (t) + h(t) so we get E = span(H, f1 ).

b – We have clearly that H0 ∪ {f1 } ∈ E0 , so span(H0 , f1 ) ⊂ E0 . Let f ∈ E0 , we

consider f in the following form f (t) = λf1 (t) + h(t), with h ∈ H0 , we have f (1) =
λf1 (1) + h(1)+ = λ/2 then λ = 2f (1) so we get f (t) = 2f (1)f1 (t) + h(t) we get that
h ∈ H, now we prove that h ∈ E0 , h(t) = f (t) − 2f (1)f1 (t) then
Z 1 Z 1 Z 1
f (t) − 2f (1)f1 (t)dt = f (t)dt − 2f (1) f1 (t)dt
0 0 0
1 2 1
 t=1
= −2f (1) t − t =0
2 2 t=0

thus h ∈ E so E0 = span(H0 , f ).

c – Consider the sequence {fn (t)}n∈N ⊂ H0 defined as follows

 



1 − 1
2 n t t ∈ [0, 1/n]
∀n ∈ N, fn (t) = t− 1
2 t ∈ [1/n, 1 − 1/n]
 
 1 − 1n t + 1n − 1

t ∈ [1 − 1/n, 1]
2 2

we have that

– ∀n ∈ N, fn ∈ H0 : let n ∈ N, fn (1) = (1 − 21 )1 + 1
2 − 1 = 0 so fn ∈ H. And also

Z 1 Z 1/n Z 1−1/n Z 1
fn (t)dt = fn (t)dt + fn (t)dt + fn (t)dt = 0
0 0 1/n 1−1/n

– limn→∞ fn = f1 : we have

3
Z 1 Z 1/n
||fn − f1 ||2 = fn (t) − f1 (t)dt = 2 fn (t) − f1 (t)dt = →0
0 0 2n

thus fn converges to f1 and we have that f1 ∈ H0 .

Question (3). Let g ∈ H0⊥ be arbitrary

R1
a) Show that 0 f1 (t)g(t)dt ≤ ||f1 − fn ||2 ||g||2 .
R1
b) Deduce that 0 f1 (t)g(t)dt ≤ √1 ||g||2 ,
n
and that g ∈ E0⊥ .

R1 R1 2
c) Show that 0 |g(t)|2 dt = 0 g(t)dt .

d) Deduce that g ≡ λ and that g ≡ 0.

< ;8< ;
6
< ;

a – We have the following

Z 1
f1 (t)g(t) = |⟨f1 , g⟩| = |⟨f1 − fn + fn , g⟩|
0
= |⟨f1 − fn , g⟩ + ⟨fn , g⟩|
= |⟨f1 − fn , g⟩|
≤ ||f1 − fn ||2 ||g||2

b – From before we have that |⟨f1 , g⟩| ≤ ||f1 − fn ||2 ||g||2 we have
3 1
||f1 − fn || = ≤√
2n n
for n ≥ 2 then we have that |⟨f1 , g⟩| ≤ √1 ||g||2 .
n
Taking n → ∞ we see that
⟨f1 , g⟩ = 0 then g ⊥ f1 and also {g} ⊥ H0 so g ∈ E0⊥ .

c – By taking m = 01 g(t)dt then we have that ⟨m − g, g⟩ = 0 so ⟨m, g⟩ = ⟨g, g⟩ = ||g||22

R

thus by the Cauchy-Schwarz inequality we have

Z 1 Z 1 Z 1 2
|g(t)| dt =
2
||g||22 = ⟨m, g⟩ = m g(t)dt = g(t)dt
0 0 0

d – We have from the previous quesiton that ||g||22 = |⟨g, 1⟩|2 which is the Cauchy-
Schwarz inequality when it is attained. And we have that it is attained only when
g ≡ λ1 then ∀t ∈ [0, 1], g(t) = λ. Given that g is in the orthogonal of H0 in H, then
g is in H so g(1) = λ = 0 thus g ≡ 0.

Question (4). Deduce that H is not a Hilbert space.

We have that H0 is closed in H and H0⊥ = {0} thus H0 is dense in H, then it should be
that H0 = H but it is not the case, thus H is not a Hilbert space.

Problem 3
Consider H the Hilbert space l2 .

Question (1). Show that ∀{yn }n∈N ∈ H, ∀a > 0, there exists infinitely many
terms of {yn }n∈N such that |yn |2 ≤ a/n.

Let {yn }n∈N ∈ H and a > 0, we want to prove that ∀N ∈ N, ∃n ≥ N, |yn |2 ≤ a/n. Suppose
by contradiction the inverse, that is ∃N ∈ N, ∀n ≥ N, |yn |2 ≥ a/n. We have then
∞ ∞ ∞
a 1
=a
X X X
2
|yn | ≥
k=N k=N
n k=N
n

which means that {yn }n∈N ̸∈ l2 which is a contradiction, thus there exists infinitely many
terms of {yn }n∈N such that |yn |2 ≤ a/n.
< ;8< ;
7
< ;

Question (2). Let V be a neighborhood of 0 in the topology σ(H, H ′ ), then

there exists x1 , . . . , xN ∈ H such that V = {x ∈ H | ∀i ∈ 1, n, |⟨x, xi ⟩| < ϵ}

We have from the weak topology that the system of neighborhoods of 0 is written in the
form

V = {x ∈ H | ∀i ∈ 1, n, |⟨fi , x − 0⟩| ≤ ϵ}

where f1 , . . . , fN ∈ H ′ . By Riesz reprenstation theorem, we have that ∃x1 , . . . , xN ∈ H

such that ∀i ∈ 1, n, ∀x ∈ H, ⟨fi , x⟩ = ⟨xi , x⟩. And since |⟨x, xi ⟩| = |⟨x, xi ⟩| = ⟨xi , x⟩, then
we obtain the form in the question directly.

Question (3). Consider z = {zn }n∈N such that ∀n ∈ N, zn = k=1 |(xk )n |.

PN

Prove that {zn }n∈N ∈ H.

By evaluating the sum of squares of {zn }n∈N we have

N 2 N
N X
|zn | = (xk )n = (xk )n (xl )n
X X X X X
2

n≥0 n≥0 k=1 n≥0 k=1 l=1

N X
N N X
= |(xk )n (xl )n | = |(xk )n | · |(xl )n |
XX X

n≥0 k=1 l=1 l,k=1 n≥0

 1/2  1/2
N N
|(xk )n | |(xl )n | =
X X X X
2 2
≤  · ||xk ||22 · ||xl ||22
k,l=1 n≥0 n≥0 k,l=1

All the terms are positive and the sum is finite then the series n≥0 |zn |
2 converges so
P

{zn }n≥N ∈ H.

√
Question (4). Prove ∀k ∈ 1, N , ∃M ∈ N, ∀n ≥ M, |(xk )n | ≤ ϵ/ n.

We have that |(xk )n |2 ≤ ϵ2 /n, we have also from the previous question that zn ∈ l2
√
then n≥0 |(xk )n |2 converges thus ∃n0 ∈ N such that ∀n ≥ n0 satisfying |(xk )n | ≤ ϵ/ n
P

otherwise the seires n≥0 |(xk )n |2 is greater than the series n≥0 1/n.
P P

√
Question (5). Deduce that V ∩ { nen , n ∈ N} =
̸ ∅.

From question (4), we know that for any k ∈ {1, 2, . . . , N }, |xk (n)| ≤ √εn except for a finite
√
number of indices. This implies that for large enough n, the sequence nen will satisfy
√
the condition |⟨ nen , xk ⟩| ≤ ε for all k. Since V is defined as the set of all x ∈ H such
√
that |⟨x, xk ⟩| ≤ ε for all k, it follows that nen ∈ V for sufficiently large n. Therefore,
√
V ∩ { nen , n ≥ 1} = ̸ ∅.

< ;8< ;
8
< ;

Question (6). Conclude.

The conclusion from question (5) is that any neighborhood V of 0 in the weak topology
√
σ(H, H ′ ) contains elements of the form nen for infinitely many n. Then 0 is an adherent
√
point to the sequence nen , but it does not converge to it.

√
Question (7). Is there any weakly convergent subsequence of nen to 0.

√ √
Suppose by contradiction that nen has a subsequence nk enk that weakly converges to
√
0, by the Banach-Stainhause theorem we have that nk enk should be bounded, but it is
√
not the case since limk→∞ || nk enk || = ∞. Thus we get our contradiction, so it does not
contain any weakly 0 convergent subsequence.

Problem 4
Method 1

Let E be a Banach space of infinite dimension. We will show that the weak topology is not
metrizable. Suppose to the contrary that d is a distance such that its induced topology is
σ(E, E ′ ). ).

Question (1). Show that ∀n ∈ N, ∃ϵn > 0 and a finite Fn ⊂ E ′ such that
any neighborhood of 0 contain a set of the form

Vn = {x ∈ E | |l(x)| ≤ ϵn , ∀l ∈ Fn }

Let us show that for any n ∈ N, there exist εn > 0 and a finite set Fn ⊂ E ′ such that any
neighborhood of 0 contains a set Vn of the form

Vn = {x ∈ E : |l(x)| ≤ εn , ∀l ∈ Fn }

We have by assumption the weak topology σ(E, E ′ ) is metrizable. Therefore, there exists
a countable base of neighborhoods of 0. Let {Vn }n∈N be such a base. For each n ∈ N,
there exists εn > 0 and a finite set Fn ⊂ E ′ such that

Vn = {x ∈ E : |l(x)| ≤ εn , ∀l ∈ Fn }

and since {Vn }n∈N is a neighborhood basis then for any neighborhood U of 0 the exist
some n ∈ N such that: Vn ⊂ U , as required.

Question (2). Let Yn = span(Fn ). Show the existance of l0 ∈ E ′ and

l0 ̸∈ ∪k≥1 Yk .

We have Yn is proper finite-dimensional space, so it is closed in the weak topology and it

interior is empty. Consider the union n≥1 Yn . By Baire’s category theorem, since E ′ is a
S

< ;8< ;
9
< ;

Banach space and hence a complete metric space, the countable union of closed sets with
empty interior is with empty interior so it cannot cover E ′ . Therefore, there exists some
l0 ∈ E ′ such that l0 ∈
/ n≥1 Yn .
S

Question (3). The goal is to show that U = {x ∈ E | |l0 (x)| ≤ 1} does not
contain any of Vn .

a) Show that if ϕ, ϕ1 , . . . , ϕn ∈ E ′ , ∩1≤k≤n ker ϕk ⊂ ker ϕ then ϕ = λ k ϕk .

P

b) Show that l0 =
P
k λk lk

Next, we want to show that the set U = {x ∈ E : |l0 (x)| ≤ 1} does not contain any of the
Vn .

a) Consider the map F : X → Rk+1 defined by

F (u) = φ(u), φ1 (u), φ2 (u), . . . , φk (u) .
 

It follows from assumption (2) that a = [1, 0, 0, . . . , 0] does not belong to R(F ).
Thus, one can strictly separate {a} and ImF by some hyperplane in Rk+1 ; i.e.,
there exist constants µ, µ1 , µ2 , . . . , µk and α such that
k
µ < α < µφ(u) + µi φi (u)
X
∀u ∈ E.
i=1
It follows that
k
µφ(u) + µi φi (u) = 0 ∀u ∈ X,
X

i=1
and also µ < 0 (so that µ ̸= 0).

k
µi
φ= with λi =
X
λi φi , .
i=1
µ
Suppose, for contradiction, that there exists n ∈ N such that the set
Vn = {x ∈ E : |l(x)| ≤ εn , ∀l ∈ Fn }
is a subset of U , where U = {x ∈ E : |l0 (x)| ≤ 1}. We will show that this assumption
leads to the conclusion that l0 is a linear combination of the functionals in Fn , which
contradicts the construction of l0 .

If Vn ⊂ U , then for every x ∈ Vn , we have |l0 (x)| ≤ 1. In particular, for any x such
that l(x) = 0 for all l ∈ Fn , we must also have l0 (x) = 0. This means:
ker l ⊂ ker l0 .
\

l∈Fn

Applying the previous results we get that l0 is a linear combination of the functionals
in Fn . That is, there exist scalars λk ∈ K such that:
mn
l0 =
X
λk lk .
k=1
< ;8< ;
10
< ;

However, by construction, l0 ∈ / ∞ n=1 Yn , where Yn = Span(Fn ). This means l0

S

cannot be expressed as a linear combination of any finite subset of F = {ln,k : n ∈

N, k = 1, . . . , mn }. and thus we get our contradiction, and so for all n ∈ N, Vn ̸⊂ U

Question (4). What can we conclude ?

We conclude that the weak topology does not have a countable neighborhood basis at 0,
and hence it is not metrizable.

Method 2

σ(E,E ′ )
Question (1). Show that ∀n ∈ N, nS
f = nS = B(0, n)

Let x ∈ E such that ||x|| < n and V a neighborhood of x, we have that V = {y ∈

E | ∀i ∈ I, |⟨fi , y − x⟩| < ϵ} for finite I and {fi }i∈I ⊂ E ′ . Fix y0 ∈ E \ {0} such that
∀i ∈ I, ⟨fi , y0 ⟩ = 0 which exists since if it does not we get that ϕ(x) = (fi (x))i∈I is an
isomorphism between E and a finite dimensional space which is not the case. Take the map
g(t) = ||x + ty0 || then we have that g(0) = ||x|| < n and that limt→∞ g(t) = ∞ then there
exists a value t0 such that g(t0 ) = ||x + t0 y0 || = n. Which gives that x + t0 y0 ∈ V ∩ nS
which concludes our proof given that B(0, n) is closed in the weak topology since it is
convex and it is closed in the strong topology.

Question (2-3). Suppose that σ(E, E ′ ) is defined by a distance d. Show

that there exists a sequence {xn }n∈N such that ∀n ∈ N, ||xn || = n and
d(xn , 0) ≤ n. Conclude.

Given that by assumption we took that σ(E, E ′ ) is defined by a distance d then it is second
countable, so for set A ⊂ E, for any element in A, there exists a sequence that converges
to it. We consider the sequence {xn }n∈N ⊂ E with ||xn || = n such that xn → 0 since
f = B(0, n). But we get that limn→∞ ||xn || = ∞ so xn does not converge to 0. Thus
nS
σ(E, E ′ ) is not metrizable.

< ;8< ;
11