You are on page 1of 5

D-MATH Functional Analysis I HS 2012

Sample Solutions for Exercise Sheet 3

R
1. (a) Let X be a normed vector space and F : X → a non-zero linear map. Show that the
kernel of F is either closed or dense in X. Also show that F is continuous if and only if the
kernel of F is closed.

F induces an isomorphism T between X/ ker F and R


because two elements of X have
the same image in Rif and only if their difference lies in the kernel of the linear map F .
Thus dim(X/ ker F ) = 1. Note that both ker F and ker F are linear subspaces of X. Since
ker F ⊂ ker F , we must therefore either have ker F = ker F or ker F = X.

Let F be continuous. Then ker F = F −1 ({0}) must be closed since it is the preimage of a
closed set.
Now let ker F be closed. Let Q : X → X/ ker F be the natural quotient map. By Satz 2.3.1
of the lecture notes, Q is continuous (since ker F is closed). Let T be as above. Hence T is
continuous since X/ ker F is finite-dimensional (see Satz 2.2.2 of the lecture notes). Now,
F = T ◦ Q by the defintion of T . Hence, F is continuous as a composition of continuous
maps.

(b) Find a normed space X with dim X = ∞, a dense subset D ⊂ X and a linear map
R
F : X → such that F is bounded on D but not continuous on X.

Let X = C 1 ([0, 1]) be the space of real-valued smooth functions on the interval [0, 1] with the
uniform norm, i.e.
kf k = sup |f (x)|
x∈[0,1]

R
for all f ∈ X. Let F : X → , f 7→ f 0 (0). This map is linear by the properties of the
derivative. However, it is not continuous on X: Consider the sequence f = (fn ) with

sin(n2 x)
fn (x) =
n
in X. This sequence converges uniformly to the zero function, but

n2 cos(n2 · 0)
F (fn ) = = n → ∞.
n
However, fn → 0 as n → ∞.
Set
D = ker F = {f ∈ C 1 ([0, 1]) | f 0 (0) = 0}.
F is of course bounded on D and D is dense in X by (a). Thus, all requirements are fulfilled.

1
2. (a) Let 1 ≤ p < ∞. Show that the map

T1 : C([0, 1]) → R, f 7→ f (1)

is not continuous if C([0, 1]) is endowed with the topology which is induced by the norm k · kp
R
and is endowed with the usual topology.

Recall that a linear operator T : X → Y between normed spaces is continuous if and only if
it is bounded, i.e. if and only if supx6=0 kT xkY
kxkX exists (cf. §2.2 of the lecture notes).

Consider the sequence of continuous functions

R, x 7→ xn
fn : [0, 1] → (n ∈ N).
Then T1 (fn ) = fn (1) = 1 for all n ∈ N and
Z 1 1/p  1/p
np 1
kfn kp = x dx = .
0 np + 1
1 (fn )kR
Thus, kTkf n kp
= (np + 1)1/p which is unbounded as n → ∞. By the remark above, this
shows that T is not continuous.

(b) Let a, b ∈ R with a < b. Show that the map


R,
Z b
2
T2 : L ([a, b]) → f 7→ f (x) dx
a

is well-defined, linear and continuous. Compute its norm.

The Cauchy-Schwarz inequality implies that the set of functions in L2 ([a, b]) is a subset of
the set of functions in L1 ([a, b]) (see the proof of Exercise 2 on Exercise Sheet 2). Thus, any
function in L2 ([a, b]) is Lebesgue-integrable on the interval [a, b] which shows that T2 is well-
defined. The linearity of the Lebesgue integral implies the linearity of T2 . By the remark above,
it remains to show that T2 is bounded. For all f ∈ L2 ([a, b]), we have
Z b Z b Z b

|T2 (f )| =
f (x) dx ≤
|f (x)| dx = |1 · f (x)| dx
a a a
Z b 1/2 Z b 1/2
≤ 1 dx |f (x)|2 dx = (b − a)1/2 kf k2 ;
a a

the second inequality follows here from the Cauchy-Schwarz inequality. Thus,

|T2 (f )| √
kT2 k = sup ≤ b−a (1)
f ∈L2 ([a,b])r{0} kf k2

which proves the continuity of T2 .


Let
h : [a, b] → R, x 7→ √
1
b−a
.

2

Then h ∈ L2 ([a, b]) and khk2 = 1 and T2 (h) = b − a. Thus,
√ |T2 (h)| (1) √
b−a= ≤ kT2 k ≤ b − a,
khk2

hence kT2 k = b − a.

(c) Show that the map

T3 : `2 → `2 , (a1 , a2 , a3 , . . .) 7→ (a2 , a3 , a4 , . . .)

is a linear and continuous operator. Show that its norm equals 1, but T3 is not an isometry.

The linearity of the operator T3 is obvious. Let a = (a1 , a2 , . . .) ∈ `2 . We have


X X
kT3 (a)k22 = |an |2 ≤ |an |2 = kak22 .
n≥2 n≥1

Thus kT3 k ≤ 1. Taking e(2) = (0, 1, 0, 0, . . .), we have T3 (e(2) ) = e(1) = (1, 0, 0, . . .), hence
ke(2) k2 = kT3 (e(2) )k2 and therefore 1 ≤ kT3 k. Thus kT3 k = 1.

Recall: A linear map T : X → Y between two normed spaces is called an isometry if


kT xkY = kxkX for all x ∈ X.
We have T3 (e(1) ) = 0 = (0, 0, . . .). Hence, ke(1) k = 1 6= 0 = k0k = T3 (e(1) ) which shows
that T3 is not an isometry.

3. Let X be a compact metric space and let F : X × X → [0, 1] be a continuous map. For every
y ∈ X we consider the map

fy : X → [0, 1], x 7→ F (x, y).

Show that the set of functions {fy | y ∈ X} is equicontinuous.

Let d denote the metric on X. The product topology on X × X is induced by the metric
p
dX×X ((x1 , y1 ), (x2 , y2 )) = d(x1 , x2 )2 + d(y1 , y2 )2 .

Moreover, the product space X × X is compact since X is compact (cf. Theorem 26.7 in
Munkres, Topology). By the Heine-Cantor theorem, F is uniformly continuous. Hence, for
each ε > 0 there exists δ > 0 such that

dX×X ((x1 , y1 ), (x2 , y2 )) < δ ⇒ |F (x1 , y2 ) − F (x2 , y2 )| < ε

for all x1 , x2 , y1 , y2 ∈ X. But this implies that for each ε > 0 there exists δ > 0 such that for
all x1 , x2 ∈ X and all y ∈ Y , we have

d(x1 , x2 ) = dX×X ((x1 , y), (x2 , y)) < δ ⇒ |fy (x1 ) − fy (x2 )| = |F (x1 , y) − F (x2 , y)| < ε

which is is exactly the condition for the equicontinuity of the family {fy | y ∈ X} (see (6.3.1)
in the lecture notes).

3
4. Let C 1 ([0, 1]) be the space of all continuously differentiable functions f : [0, 1] → . Set R
kf k := kf k∞ + kf 0 k∞ for all f ∈ C 1 ([0, 1]). Show that (C 1 ([0, 1]), k · k) is a Banach space.
Deduce that (C 1 ([0, 1]), k · k∞ ) is not a Banach space.

The function k · k∞ : C 1 ([0, 1]) → [0, ∞), f 7→ supx∈[0,1] |f (x)| clearly defines a norm on
C 1 ([0, 1]). Moreover, the map

C 1 ([0, 1]) → [0, ∞), f 7→ kf 0 k∞ = sup |f 0 (x)|


x∈[0,1]

is positive homogeneous and satisfies the triangle inequality (however, it is not a norm on
C 1 ([0, 1])!). Thus,

k · k : C 1 ([0, 1]) → [0, ∞), f 7→ kf k∞ + kf 0 k∞

is a positive homogeneous map and satisfies the triangle inequality. In order to assess that k · k
is a norm on C 1 ([0, 1]), it remains to check that kf k = 0 implies f = 0; however, this is clear
since kf k∞ = 0 implies f = 0 and kf 0 k∞ is non-negative.
We now prove the completeness of (C 1 ([0, 1]), k · k): Let (fn ) be a Cauchy sequence in
(C 1 ([0, 1]), k · k). By the positivity of k · k∞ , it follows that (fn ) and (fn0 ) are both Cauchy se-
quences in (C([0, 1]), k · k∞ ). Since (C([0, 1]), k · k∞ ) is a Banach space, there exist functions
f ∈ C([0, 1]) and g ∈ C([0, 1]) with

kfn − f k∞ → 0 and kfn0 − gk∞ → 0 as n → ∞

and we have f ∈ C 1 ([0, 1]) and g = f 0 (see Theorems 7.12 and 7.17 in Rudin, Principles of
mathematical analysis). Hence kfn − f k → 0 as n → ∞. Therefore, (C 1 ([0, 1]), k · k) is a
Banach space.
Suppose that (C 1 ([0, 1]), k · k∞ ) is a Banach space. Since we have kf k∞ ≤ kf k for all
f ∈ C 1 ([0, 1]), the Open Mapping Theorem (or rather its immediate corollary Beispiel 3.2.1
in the lecture notes) implies that the norms k · k∞ and k · k are equivalent on C 1 ([0, 1]).
Thus, there exists a constant α > 0 such that αkf k∞ ≥ kf k for all f ∈ C 1 ([0, 1]). Let
N
fn (x) := xn ∈ C 1 ([0, 1]) for all n ∈ . Then

α − 1 = (α − 1)kfn k∞ ≥ kfn k − kfn k∞ = kfn0 k∞ = n

for all n ∈ N; this is clearly false. Therefore, (C 1([0, 1]), k · k∞) is not complete.
5. Let (X, k·k) be a Banach space and let L1 , L2 be two closed, linear subspaces with L1 ∩L2 =
{0}. Show that L1 + L2 is closed if and only if there exists a constant α ≥ 0 such that

kxk ≤ αkx + yk (2)

for all x ∈ L1 and all y ∈ L2 .

Note: L = L1 + L2 is clearly a linear subspace of X.

Let L be closed in (X, k · k). Let

T : L → X, z = x + y 7→ x

4
where x ∈ L1 and y ∈ L2 ; note that this map is well-defined because of the assumption
L1 ∩ L2 = {0} of the exercise.
We claim that the graph ΓT = {(z, T z) | z ∈ L} ⊂ X × X is closed. By the Closed Graph
Theorem, T is then continuous and therefore also bounded; thus, there exists α ≥ 0 such that
kT zk kxk
α = kT k = sup ≥ (3)
z∈Lr{0} kzk kx + yk

x ∈ L1 and all y ∈ L2 with x + y 6= 0; thus, the inequality (2) holds in this case. By the
assumption L1 ∩ L2 = {0} and since L1 and L2 are linear subspaces of X, we have x + y = 0
if and only if x = y = 0; in this case, inequality (2) obviously holds as well.
It remains to prove the claim: Let ((zn , wn ))n = ((zn , T zn ))n be a sequence in ΓT which
converges to (z, w) ∈ X ×X. Then z ∈ L since L is closed by assumption. Moreover, we have
T (L) = L1 which is also closed. Thus, w ∈ L1 . Now note that zn − T zn ∈ L2 for all n ∈ N
because of the definition of T . Being the difference of two converging sequences, (zn − T zn )
converges to z − w ∈ X. Since L2 is closed, we have z − w ∈ L2 . By the definition of T and
because z ∈ L, we also have z − T z ∈ L2 . Therefore, T z − w = (z − w) − (z − T z) ∈ L2
since L2 is a linear subspace. But we also have T z − w ∈ L1 since L1 is a linear subspace.
Hence, T z − w ∈ L1 ∩ L2 = {0} and thus

lim T zn = lim wn = w = T z = T ( lim zn ),


n→∞ n→∞ n→∞

which implies that ((zn , T zn )) converges in ΓT , i.e. ΓT is closed in X × X.

Now assume that there exists α ≥ 0 such that

kxk ≤ αkx + yk

for all x ∈ L1 and all y ∈ L2 . Then, by (3), T is a bounded operator, hence continuous.
Let (zn ) be a sequence in L which converges to some z ∈ X. Since T is continuous, the
sequence (T zn ) is a converging sequence in L1 . The limit x of this sequence is in L1 since L1
N
is closed by assumption. Let yn = zn − T zn for all n ∈ ; note that yn ∈ L2 for all n ∈ byN
the the definition of T . Being the difference of two converging sequences, (yn ) ⊂ L2 converges
to z − x ∈ X. But z − x ∈ L2 since L2 is closed. Therefore, z = x + (z − x) ∈ L1 + L2 = L
which implies that L is closed.

You might also like