Functional Analysis Exam

(1) Let X := {x ∈ KN : ∃ limn→+∞ x(n)} be the space of convergent K-valued sequences,

equipped with the norm kxk∞ := supn∈N |x(n)|. Prove or disprove: (X, k k∞ ) is a Banach
space.
(5 points)
Solution: We show that (X, k k∞ ) is a Banach space. Since X is a subspace of l∞ :=
l∞ (N, K) (equipped with the same norm), and l∞ is a Banach space, it is sufficient to show
that X is closed in l∞ . Let xn , n ∈ N, be a sequence in X, converging to some x ∈ l∞ . For
an ε > 0, let n ∈ N be such that kx − xn k∞ < ε/3. Since xn is convergent, it is also Cauchy,
and therefore there exists Nε such that |xn (k) − xn (l)| < ε/3 for all k, l ≥ Nε . Thus

|x(k) − x(l)| = |x(k) − xn (k) + xn (k) − xn (l) + xn (l) − x(l)|

≤ |x(k) − xn (k)| + |xn (k) − xn (l)| + |xn (l) − x(l)|
< kx − xn k∞ + ε/3 + kx − xn k∞ < ε,

for all k, l ≥ Nε . Hence, x is a Cauchy sequence in K, and therefore convergent, i.e., x ∈ X.

This proves the assertion.

(2) Equip X := C([0, 1], K) with the norm kf k∞ := maxt∈[0,1] |f (t)|. Prove that there exists no
inner product h., .i : X × X → K such that kf k2∞ = hf, f i for all f ∈ X.
(4 points)
Solution: Assume the contrary; then we have the parallelogram law, i.e., for every f, g ∈ X,
kf + gk2 + kf − gk2 = 2(kf k2 + kgk2 ). Choose f, g ∈ X such that they are supported on
disjoint sets, and kf k∞ = kgk∞ = 1. Then kf + gk = kf − gk = 1, and hence

kf + gk2 + kf − gk2 = 2 < 4 = 2(kf k2 + kgk2 ),

a contradiction.

(3) Let X be a normed space and A ⊆ X an arbitrary subset of X. Show that A is bounded if
and only if for every f ∈ X ∗ it holds that sup{|f (x)| : x ∈ A} < ∞.
(1+4 points)
Solution: If A is bounded then there exists an M > 0 such that kxk ≤ M for all x ∈ A, and
hence for every f ∈ X ∗ , sup{|f (x)| : x ∈ A} ≤ kf k M < ∞.
Assume now that sup{|f (x)| : x ∈ A} < ∞ for all f ∈ X ∗ . Consider the canonical
embedding of X into X ∗∗ , denoted by J : X → X ∗∗ ; then kJ(x)k = kxk, x ∈ X. By
assumption we have that sup{|J(x)(f )| : x ∈ A} = sup{|f (x)| : x ∈ A} < ∞ for all f ∈ X ∗ .
As X ∗ is a Banach space, we can apply the Banach-Steinhaus theorem to the set {J(x) :
x ∈ A} ⊂ B(X ∗ , K) = X ∗∗ to get that sup{kxk : x ∈ A} = sup{kJ(x)k : x ∈ A} < ∞.

1
(4) Let T be a linear operator on a Hilbert space H such that

hT x, yi = i hx, T yi , ∀ x, y ∈ H, (0.1)
√
where i = −1 is the imaginary unit. Show that T is bounded.
(4 points)
Solution: By the closed graph theorem, it is enough to show that T is closed, which in turn
is equivalent to the following: if limn→+∞ (xn , T xn ) = (0, y) then y = 0. But we have
D E
kyk2 = hy, yi = lim T xn , y = lim hT xn , yi
n n
D E
= lim i hxn , T yi = i lim xn , T y = i h0, T yi = 0,
n n

where in the 3rd and the 5th equality we used the continuity of the inner product, and in
the 4th equality the assumption (0.1). This proves the assertion.

(5) Let X := Y := c00 := x ∈ KN : #{n : x(n) 6= 0} < +∞ be the space of sequences that
take zero value at all but finitely many indeces. Recall that the p-norms are defined on c00
p 1/p
P

as kxkp := n∈N |x(n)| for all 1 ≤ p < +∞.
a) Compute the norm of the left shift (Sx)(n) := x(n + 1), x ∈ c00 , n ∈ N, as an operator
from (X, k k3 ) to (Y, k k5 ). b) Compute the norm of the same operator when Y is equipped
with k k1 instead.
(4+4 points)
Solution: a) We have seen that k kq ≤ k kp whenever 1 ≤ p ≤ q ≤ +∞. Hence,

+∞
!1/3
X
kSxk5 ≤ kSxk3 = |x(n)|3 ≤ kxk3 ,
n=2

and hence kSk ≤ 1. On the other hand, if ek (n) := δk,n is the indicator function of {k} with
k ≥ 2 then kxk3 = 1, and kSxk5 = kek−1 k5 = 1, and hence kSk ≥ 1. Combining the two
inequalities, we get kSk = 1.
b) For every N ∈ N, let xN := N
P +1
k=2 ek = (0, 1, 1, . . . , 1, 0, 0 . . .), where after an initial 0 we
have N consecutive 1’s, and then all 0’s. Then kxN k3 = N 1/3 , while kSxN k1 = N . Hence,

kSxN k1
kSk ≥ sup = sup N 2/3 = +∞.
N ∈N kx k
N 3 N ∈N

That is, S is unbounded.

R1
(6) Consider H := L2 ([0, 1], C), equippped with its usual inner product hf, gi := 0 f (t)g(t) dt,
R1
and define T : H → H as (T f )(x) := 0 x2 yf (y) dy, f ∈ H, x ∈ [0, 1]. Compute the norm
of T . Compute the rank of T , i.e., dim ran T . Is T compact?
(Norm: 4 points, rank: 1 point, compactness: 1 point)

2
 Solution: T is of the form T (f ) = α hf, βi, where α(t) = t2 , β(t) = t, t ∈ [0, 1]. Hence, for
every f ∈ H, kT k = | hf, βi | kαk ≤ kf k kβk kαk, where we have used the Cauchy-Schwarz
inequality. Moreover, the inequality holds with
2 R 1 4equality if we choose
2 Rf 1:=2 β/ kβk, and hence
we have kT k = kβk kαk. Since kαk = 0 t dt = 1/5, kβk = 0 t dt = 1/3, we get
√
kT k = 1/ 15.
It is clear from T (f ) = α hf, βi and β 6= 0 that ran T is the one-dimensional subspace
spanned by α, and hence the rank of T is 1. Since every finite-rank operator is compact, T
is compact.

(7) Let X := C([0, 1], C) be equipped with kf k∞ := supt∈[0,1] |f (t)|, and define M : X → X by
(M f )(t) := tf (t), f ∈ X, t ∈ [0, 1]. Prove that the spectrum of M is σ(M ) = [0, 1], and
determine the point spectrum, the residual spectrum, and the continuous spectrum of M .
(3+1+3+1 points)
Solution:
1
Assume that λ ∈ C \ [0, 1], and define (Rλ f )(t) := λ−t f (t), f ∈ X, t ∈ [0, 1]. By as-
1
sumption, inf t∈[0,1] |λ − t| > 0, and hence Cλ := supt∈[0,1] |λ−t| < +∞. Thus, kRλ f k∞ =
 1 
supt∈[0,1]  λ−t f (t) ≤ Cλ kf k∞ , and therefore kRλ k ≤ Cλ , i.e., Rλ is bounded. Obviously,
(λI − M )Rλ = Rλ (λI − M ) = I, i.e., Rλ is the inverse of λI − M , and therefore λ is in the
resolvent set of M . Thus, σ(M ) ⊆ [0, 1].
Now let λ ∈ [0, 1]. Assume that 0 = (λI − M )f for some f ∈ X, i.e., (λ − t)f (t) = 0 for all
t ∈ [0, 1]. This implies that f (t) = 0 for all t ∈ [0, 1] \ {λ}, and by the the continuity of f ,
we also have f (λ) = 0. Thus, f is the zero function, and hence it cannot be an eigenvector.
This shows that σp (M ) = ∅. Moreover,

ran(λI − M ) ⊆ {f ∈ X : f (λ) = 0} =: Xλ .

Obviously, Xλ is a subspace that is not equal to X, and it is easily seen to be closed; indeed, if
limn→+∞ kfn − f k∞ = 0 and fn (λ) = 0 for all n ∈ N then |f (λ)| ≤ |f (λ) − fn (λ)| + |fn (λ)| =
|f (λ) − fn (λ)| ≤ kf − fn k∞ −−−−→ 0 yields that f (λ) = 0. Hence, ker(λI − M ) = {0} and
n→+∞
ran(λI − M ) 6= X, and thus λ ∈ σr (M ). Thus, σr (M ) = [0, 1] = σ(M ), which also gives
σc (M ) ⊆ σ(M ) \ σr (M ) = ∅, i.e., σc (M ) = ∅.

Grand total: 40 points