FUNCTIONAL ANALYSIS - IMPA (2023)

SOLUTIONS OF EXAM 1

Problem 1. Let E be a normed vector space over K.

(i) (3 points) Prove that E is separable if its dual space E ∗ is separable.
(ii) (2 points) Prove or present a counter-example: If E is separable, then E ∗ is separable.

Solution.
(i) Let {fn }n be a dense set in the close unit ball B ⊂ E ∗ . Assume that E is not separable. This
implies that given any countable set {xn } ⊂ E, the set Y = (span){xn } =
̸ E. By the second
geometric form of Hahn-Banach Theorem, we can produce a function f ∈ E ∗ such that f (y) = 0
for every y ∈ Y and ∥f ∥ = 1. By density, and changing the order if necessary, this means we can
assume that ∥f − f1 ∥ < 1/4. This would mean

|f (x1 )| = |f1 (x1 ) − (f (x1 ) − f (x1 ))|

≥ |fn (x1 )| − ∥f − f1 ∥∥x1 ∥

Since ∥f − f1 | < 1/4 and ∥f ∥ = 1, this means ∥f1 ∥ > 3/4. Therefore we can choose x1 to be such
that ∥xn ∥ = 1 and fn (xn ) > 1/2. This would mean

|f (xn )| > 1/4

But f = 0 in Y , which is absurd. Therefore, we have that E is separable.

(ii) Counter-example: Choose E = ℓ1 . E is a separable, but its dual is isometric to ℓ∞ , which is not
separable.

1
Problem 2. Let E and F be Banach spaces over K.

(i) (1 point) Prove or present a counter-example: A linear operator T : E → F such that Ker(T ) =
{x ∈ E : T x = 0} is a closed subspace is bounded.

(ii) (2 points) Let G ⊂ E be a closed subspace. We will write x ∼ y if x, y ∈ E and x − y ∈ G. We

define E/G to be the set of all equivalence classes [x] = {y ∈ E : y ∼ x}, for all x ∈ E. Prove
that E/G is vector space over K with the operations [x] + [y] = [x + y], and λ[x] = [λx], and that

∥[x]∥E/G = inf{∥x − y∥ : y ∈ G}

defines a norm on E/G. Finally, prove this norm makes E/G a Banach space.

(iii) (2 points) Let T : E → F be a surjective bounded linear operator. Prove that the operator
Tb : E/Ker(T) → F given by

Tb([x]) = T y, for any y ∼ x,

is well defined, injective, bounded and ∥Tb∥ = ∥T ∥.

Solution.

(i) Counter-example: Let E be a infinite dimensional Banach space and let B = {ej }j∈J be a Hamel
basis of E such that ∥ej ∥ = 1 for every j ∈ J. Choose a countable infinite set I = {i1 , i2 , · · · } ⊂ J
and define T : E → E by
(
nein if j = in ∈ I
T ej =
ej otherwise.
Then Ker(T ) = {0} and T is unbounded.

(ii) The property that E/G forms a vector space are easy to check. Let us prove that ∥ · ∥E/G defines
a norm.

• If x ∼ y, then for any w ∈ G, w − x + y ∈ G and

∥x − w∥ = ∥y − (w − x + y)∥,

which means that the infimum in the definition of ∥∥E/G is the same for x and y and therefore
it is well defined and obviously non negative.

• Clearly ∥0∥ = 0.

• Choose x ∈ E such that ∥[x]∥ = 0. This means there is a sequence xn ∈ G such that
1
∥x − xn ∥ ≤ n.

Which means xn → x, and since G is closed, this means x ∈ G, and therefore [x] = [0].

• Choose λ ∈ K. If λ = 0, trivially ∥λ[x]∥ = 0 = |λ|∥[x]∥. When λ ̸= 0,

∥λ[x]∥ = ∥[λx]∥
= inf{∥λx − w∥ : w ∈ G}
= inf{|λ|∥x − λ−1 w∥ : w ∈ G}
= |λ| inf{∥x − w∥ : w ∈ G}
= |λ|∥[x]∥.

2
 • Finally, if x, y ∈ E, then

∥[x + y]∥ = inf{∥x + y − w∥ : w ∈ G}

= inf{∥x + y − 2w∥ : w ∈ G}
≤ inf{∥(x − w)∥ + ∥(y − w)∥ : w ∈ G}
≤ inf{∥(x − w)∥ : w ∈ G} + inf{∥(y − w)∥ : w ∈ G}
= ∥[x]∥ + ∥[y]∥.

We can conclude that ∥ · ∥E/G defines a norm.

Let us prove that E/G is a Banach space and for that, we will prove that every absolutely
summable sequence is convergent. Choose a sequence ([xn ])n in E/G such that
X
∥[xn ]∥ < ∞.
n

This means is a sequence wn ∈ G such that

∥xn − wn ∥ ≤ ∥[xn ]∥ + 2−n ,

Which in turn implies

X
∥xn − wn ∥ < ∞.
Since E is a Banach space, this means xn − wn is summable, i.e., there is a vector v such that
N
X
x = lim (xn − wn )
N →∞
n=1

N
X
Since wn ∈ G, one has
n=1

N
X N
X N
X
∥[x] − [xn ]∥ ≤ ∥x − xn + wn ∥
n=1 n=1 n=1
N
X
= ∥x − (xn − wn )∥
n=1

Hence
N
X
[x] = lim [xn ].
N →∞
n=1
This implies E/G is a Banach space.
(iii) It is routine to check that Tb is well defined and injective. Let us focus on proving it is bounded
and ∥Tb∥ = ∥T ∥. For every w ∈ Ker(T ), one has

∥Tb[x]∥ = ∥T (x − w)∥ ≤ ∥T ∥∥x − w∥

so by taking the infimum over w ∈ G, it follows that Tb is bounded and ∥Tb∥ ≤ ∥T ∥. On the other
hand, since ∥[x]∥E/G ≤ ∥x∥E ,

∥T x∥ = ∥Tb[x]∥ ≤ ∥Tb∥∥[x]∥ ≤ ∥Tb∥∥[x]∥∥x∥

and therefore ∥T ∥ ≤ ∥Tb∥∥[x]∥, which implies the desired result.

3
Problem 3. Given 1 ≤ p < ∞, let ℓp = ℓp (N, K) be the space of all p-summable sequences of KN , i.e,
∞
X
ℓp (N, K) = {a = (an )n ∈ KN : |an |p < ∞},
n=1

equipped with the norm

∞
!1/p
X
p
∥a∥p = |an | .
n=1
Furthermore, denote by ℓ∞ = ℓ∞ (N, K) the space of K-valued bounded sequences with the norm

∥a∥∞ = sup |an |.

n
p q
(i) (1 point) Prove that ℓ ⊂ ℓ if 1 ≤ p ≤ q ≤ ∞.

(ii) (1 point) Given a = (an )n ∈ KN , define its distribution function γ = γa : (0, ∞) → [0, ∞] by

γ(t) := #{n ∈ N : |an | > t}.

Prove that for every t > 0 and a, b ∈ KN one has

γa+b (t) ≤ γa (t/2) + γb (t/2).

(iii) (1 point) Prove Markov’s inequality: when a ∈ ℓp (N) for 1 ≤ p < ∞ and t > 0, then

γa (t) ≤ t−p ∥a∥pp .

(iv) (2 points) Let 1 ≤ p < ∞ and a ∈ ℓp . Prove that g(t) = γ(t)tp−1 is Lebesgue integrable and
Z ∞
p
∥a∥p = p γ(t)tp−1 dt.
0

Solution:

(i) For any 1 ≤ p ≤ q ≤ ∞ one has

∥a∥∞ ≤ ∥a∥q .

Furthermore, if q < ∞
X
∥a∥qq ≤ ∥a∥q−p
∞ |an |
p

= ∥a∥q−p p
∞ ∥a∥p

≤ ∥a∥qp .

Combining these inequalities implies the desired inclusion.

(ii) By triangle inequality, one has

{n : |an | ≤ t/2} ∩ {n : |bn | ≤ t/2} ⊂ {n : |an + bn | ≤ t},

Since (A ∩ B)c = Ac ∪ B c , one can take complements and obtain

{n : |an | > t/2} ∪ {n : |bn | > t/2} ⊃ {n : |an + bn | > t},

By using that #(A ∪ B) ≤ #A + #B, the desired inequality follows.

(iii) Consider (t, n) 7→ 1t (n) the indicator function of the set

{(n, t) ∈ N × (0, ∞) : |an |t−1 ≤ 1},

4
 i.e, 1t (n) = 1 if (n, t) belongs to previous set, and 0 otherwise. Clearly
X
(1) γ(t) = 1t (n)
n

It is easy to see that for any 1 ≤ p < ∞ that 1t (n) ≤ |an |p t−p . Therefore
X X
γa (t) = 1t (n) ≤ |an |p t−p = t−p ∥a∥pp .
n n

(iv) Since γ is always non-increasing, it is easy to conclude that g is Lebesgue measurable. Further-
more, by Tonelli’s Theorem and (1)
Z ∞ !
Z ∞ X
p−1
p g(t)dt = p t 1t (n) dt
0 0 n
XZ ∞
= ptp−1 1{t<|an |} dt
n 0

XZ |
= an |ptp−1 dt
n 0
X
= |an |p ,
n

which proves the result and the fact that g is Lebesgue integrable.

5
Problem 4. Let ℓp = ℓp (N, K) and γa (t) = γ(t) be defined as in Problem 3. Given 1 ≤ p < ∞, we define
the space ℓpw = ℓpw (N, K) to be set of all sequences a = (an )n ∈ Kn such that

sup{γa (t)tp : t > 0} < ∞.

We say a linear operator T : ℓp (N, K) → KN is of weak type p if there is a finite constant C > 0 such that
for every a ∈ ℓp the sequence b = T a satisfies for every t > 0 that

γb (t) ≤ Ct−p ∥a∥pp .

(i) (1 point) Prove or present a counter-example: ℓpw ⊂ ℓp , for every 1 ≤ p < ∞.

(ii) (1 point) Given a = (an )n ∈ KN and λ > 0, define aλ = (aλn )n by
(
λ an if |an | ≥ λ
an = ,
0 otherwise

and aλ = a − aλ . Assume a ∈ ℓpw for some 1 ≤ p < ∞. Prove for any λ > 0 that aλ ∈ ℓq , for
1 ≤ q < p and aλ ∈ ℓr for every p < r ≤ ∞. (In other words: ℓpw ⊂ ℓq + ℓr , when q < p < r.)
(iii) (1 point) Prove that ℓpw1 ∩ ℓpw2 ⊂ ℓp when 1 ≤ p1 < p < p2 < ∞.
(iv) (1 point) Prove that any bounded linear operator T : ℓp → ℓp is of weak type p.
(v) (1 point) Let 1 ≤ p1 < p2 < ∞ and assume that T is of weak type p1 and weak type p2 . Prove
that T defines a bounded operator from ℓp to ℓp for every p1 < p < p2 .

Solution:
(i) Counter-example: Consider an = n−p . It is easy to check that a = (an )n ∈ lw
p
\ℓp .

(ii) Since a ∈ ℓpw , there is C = Ca such that γa (t) ≤ Ct−p . In particular, the set

{n ∈ N : |an | > λ}

has cardinality smaller than Cλ−p , and therefore is a finite set. This means aλ has only finite
entries that are not zero, and therefore it belongs to ℓq for any 1 ≤ q < p. Furthermore, one has
(
γa (λ) , when t ≤ λ
γaλ =
γa (t) , when t > λ
Combining this with Problem 3 (iv), one obtains the following bound
Z ∞
∥aλ ∥qq = q γaλ (t)tq−1 dt
0
Z λ Z ∞
q−1
≤ qγa (λ) t dt + Cq tq−p−1 dt
0 λ
p
=C λq−p
p−q
As for aλ , it is clear that ∥aλ ∥ ≤ λ, so aλ ∈ ℓ∞ . Now assume p < r < ∞. It is easy to see that
(
γa (t) , when t ≤ λ
γaλ ≤
0 , when t > λ
Combining this with Problem 3 (iv), one has
Z ∞
∥aλ ∥rr = r γaλ (t)tr−1 dt
0
Z λ
≤ Cr tr−p−1 dt
0
Cr r−p
= λ ,
r−p
which means aλ ∈ ℓr , for any p < r ≤ ∞.
6
 (iii) Let 1 ≤ p1 < p < p2 < ∞. If a ∈ ℓpw2 , we know from (ii) that a1 ∈ ℓp . In the same way, from the
fact that a ∈ ℓpw1 we know that a1 ∈ ℓp . Since a = a1 + a1 , then a ∈ ℓp and we can conclude that
ℓpw1 ∩ ℓpw2 ⊂ ℓp as wanted.
(iv) By Markov’s inequality and boundedness of T , one has for any t > 0 that

γT a (t) ≤ ∥T a∥t−p ≤ ∥T ∥∥a∥t−p

therefore T is of weak type p.

(v) Let a ∈ ℓp and 1 ≤ p1 < p < p2 < ∞. Fix λ > 0 and let aλ and aλ be as in (ii). Consider 1t (n)
as in Problem 3 (iii), and define 1t (n) = 1 − 1t (n) This means
X
∥aλ ∥pp11 = |an |p1 1λ (n)
n
(2) X
∥aλ ∥pp22 = |an |p2 1λ (n).
n

By using that
|an |p1 −p 1λ (n) ≤ λp1 −p 1λ (n)
we have
X
∥aλ ∥pp11 = |an |p |an |p1 −p 1λ (n)
n

≤ λp1 −p ∥aλ ∥pp

Similarly,

∥aλ ∥pp22 ≤ λp2 −p ∥aλ ∥pp .

Since a = aλ + aλ , this means we can define T in ℓp by T a = T aλ + T aλ . By Problem 3 (iv), in

order to prove boundedness of T it is enough to have some B > 0 such for every a ∈ ℓp
Z ∞
(3) tp−1 γT a (t)dt ≤ B∥a∥pp .
0

We know from Problem 3 (ii) that

γT a (t) ≤ γT aλ (t/2) + γT aλ (t/2).

Since T is of weak type p1 and p2 , this means for some A > 0 that

γT aλ (t) ≤ A∥aλ ∥pp11 t−p1 and γT aλ (t) ≤ A∥aλ ∥pp22 t−p2 .

By choosing λ = t, we have for any δ > 0 that

Z ∞ Z ∞ Z ∞
(4) tp−1 γT a (t) dt ≤ D tp−p1 −1 ∥at ∥pp11 dt + D tp−p2 −1 ∥at ∥pp22 dt.
0 0 0

This indentity, combined with Tonelli’s Theorem, (2) and (4), yields
Z ∞ X Z ∞ X Z ∞
tp−1 γT a (t) dt ≤D |an |p1 tp−p1 −1 1t (n) dt + D |an |p2 tp−p2 −1 1t (n) dt
0 n 0 n 0

X Z |an | X Z ∞
=D |an |p1 tp−p1 −1 dt + D |an |p2 tp−p2 −1 dt
n 0 n |an |
p−p1 p−p2
X |an | X |an |
=D |an |p1 +D |an |p2
n
p − p1 n
p2 − p
1 1
By means we can obtain (3) with B = D( p−p 1
+ p2 −p ) as desired.

7
Problem 5. Let E be a Banach space over K and p : E → R be a Minkowski functional.
(i) (2 points) Assume K = R. Let F ⊂ E be a subspace and assume T : F → ℓ∞ (N, R) is a linear
operator, where ℓ∞ (N, R) and its norm are defined as in problem 3, such that for every x ∈ F

∥T x∥∞ ≤ p(x).

Prove that T can be extended to a linear operator Te : E → ℓ∞ (N, R) such that for every x ∈ E.

∥Tex∥∞ ≤ p(x).

(ii) (3 points) Let K = C and consider A, B ⊂ E to be convex subsets such that A is closed and B is
compact. Prove that there is φ ∈ E ∗ , α ∈ R and ε > 0 such that

Re{φ(x)} ≤ α − ε < α + ε ≤ Re{φ(y)}

for every x ∈ A and y ∈ B.

Solution:
(i) Consider Tn : F → K to be the n-th coordinate of the operator T . Applying Hahn-Banach
Theorem to each Tb to obtain extensions Ten : E → K satisfying

|Ten (x)| ≤ p(x).

The operator Tea = (Te1 x , Te2 x , . . .) extends T and clearly satisfies the desired bound

∥Tex∥∞ ≤ p(x)

(ii) Consider E as a vector space over R by restricting the scalar multiplication operations. A and
B retain convexity and disjointness, and they still remain closed and compact respectively. By
the second geometric form of Hahn-Banach Theorem, one can find α ∈ R and a real-valued
continuous linear functional ψ : E → R such that

ψ(x) ≤ α − ε < α + ε ≤ ψ(y)

for every x ∈ A and y ∈ B. Define γ(x) = ψ(ix). It is clear that γ defines a real-valued bounded
linear functional on E. Consider the function defined in E by

φ(x) = ψ(x) − iγ(x).

φ defines a bounded linear functional over E as a complex vector space and Re{φ} = ψ. Indeed,
the assertion about its real part being ψ is trivially true, so it remains to prove the linearity and
boundedness, hence we proceed with this task. From R-valued linearity of ψ and γ, one has

φ(x + y) = φ(x) + φ(y).

If λ = σ + it ∈ C, then by applying the R-linearity to ψ and γ, one has

φ(λx) = φ(σx + t(ix))

= σφ(x) + tφ(ix)
= σ(ψ(x) − iγ(x)) + t(ψ(ix) − iγ(i(ix)))
= σ(ψ(x) − iγ(x)) + t(γ(x) − iψ(−x))
= σ(ψ(x) − iγ(x)) + it(φ(x) − iγ(x))
= σφ(x) + itφ(x)
= λφ(x).

This means φ is linear over C. Boundedness follows directly from the boundedness of ψ and γ,
and by the observation Re{φ} = ψ, we have the desired result.
8