TECHNISCHE UNIVERSITÄT MÜNCHEN

Zentrum Mathematik
Prof. Dr. M. Wolf Funktionalanalysis Winters. 2014
M. Kech http://www-m5.ma.tum.de/Allgemeines/MA3001 2014W Lösungsblatt 4
A. Müller-Hermes (28.10.2014)

Tutoraufgaben

1. Space of Polynomials
We denote by C[z] the vector space of complex polynomials in one variable.

a) Let p = ni=0 ai xi and let kpk := ni=0 |ai |. Prove that k · k defines a norm on C[z].
P P

b) Which of the following linear functionals fj : C[z] → C are continuous in the topology
induced by k · k?
Z 1
f1 (p) := p(t)dt, f2 (p) := p0 (0), f3 (p) := p0 (1).
0

c) Which of the following linear operators Tj : C[z] → C[z] are continuous in the
topology induced by k · k?
Z z
(T1 p)(z) := p(z + 1), T2 (p)(z) := p(z 0 )dz 0 .
0

Lösung:

a) We clearly have kpk ≥ 0 for all p ∈ C[z] by construction. Furthermore from kpk = 0
we conclude that all coefficients vanish and thus p = 0. Also kαpk = |α|kpk for α ∈ C
follows directly from the construction.
Finally let us prove the triangle inequality. Let p = ki=0 pi z i and q = li=0 qi z i ,
P P
l ≥ k, then
k
X l
X
kp + qk = |pi + qi | + |qi |
i=0 i=k+1
k
X k
X l
X
≤ |pi | + |qi | + |qi | = kpk + kqk.
i=0 i=0 i=k+1

b) We have

1 k k k
|pi |
Z X X X
|f1 (p)| = | dt pi ti | ≤ | |≤ |pi | = kpk.
0 i+1
i=0 i=0 i=0

and hence f1 is continuous.

Also f2 is continuous because |f2 (p)| = |p0 (0)| = |p1 | ≤ ki=0 |pi | = kpk.
P

Consider the sequence (z n ) in C[z] and note that kz n k = 1 for all n ∈ N. We find
|f3 (z m )| = m − 1 and hence we conclude supp∈C[z],kpk=1 |f3 (p)| = ∞.
c) We find T1 (z m ) = (z + 1)m = m
P m k

m
Pm m


k=0 k z and thus kT1 (z )k ≤ k=0 k < m.
Hence T1 is not continuous.
Rz P z i+1
For p = ki=0 pi z i we find T2 (p) = 0 dt ki=0 pi ti = ki=0 pii+1
P P
. Thus, kT2 (p)k =
Pk |pi | Pk
i=0 i+1 ≤ i=0 |pi | = kpk and we conclude that T2 is continuous.
2. Finite Sequences
Consider the set of compactly supported sequences c00 := {(xn )n∈N : xn ∈ C, xi 6=
0 for non more than finitely many i}. Prove that
a) c00 6= lp for 1 ≤ p ≤ ∞.
b) c00 is dense in (lp , k · kp ) for 1 ≤ p < ∞.
c) Is c00 dense in (l∞ , k · k∞ )?
d) Prove that (c00 , k · kp ) is separable for 1 ≤ p ≤ ∞.
e) Conclude that (lp , k · kp ) is separable for 1 ≤ p < ∞.
Lösung:

a) Every Plp space contains a sequence (an ) with infinitely non-zero elements. Consider
e.g. ∞ 1
i=k k! .
b) For each sequence (xn ) in lp , 1 ≤ p < ∞, consider the sequence in c00 defined by
(xin ) = (x1 , x2 , ..., xi , 0, ...), i ∈ N.
∞
X
k(xn ) − (xkn )kp =( |xi |p )1/p
i=k+1

But for every  > 0 there is N ∈ N such that ( ∞ p 1/p < . Thus k(x ) −
P
i=N |xi | ) n
k k
(xn )kp <  for k ≥ N and we get limk→∞ (xn ) = (xn ). So for each x ∈ lp there is a
sequence in c00 that converges to x and thus c00 is dense in lp .
c) Note that x = (1, 1, 1, 1, 1, ...) ∈ l∞ . We prove that there is no sequence in c00 that
converges to x. Let (an )n∈N be a sequence in c00 . Then, for each N ∈ N there is an
i ∈ N such that the i-th element of all an , n ≤ N vanishs. Thus kan − xk∞ ≥ 1 for
all n ∈ N and hence (an ) does not converge to x.
d) Since the countable union of countable sets is countable, the set B := {(xn ) ∈ c00 :
xi ∈ Q + iQ for all n} is countable. Let x = (x1 , ..., xk , 0, ..) ∈ c00 . For each xj , each

1 ≤ p ≤ ∞ and each  > 0 there is x̃p,j ∈ Q + iQ such that |xj − x̃p,j | < k1/p because
Q is dense in R. Let x̃p = (x̃p,1 , ..., x̃p,k , 0, ...). We then find
Xk
kx − x̃p kp = ( |xj − x̃p,j |p )1/p
j=1
k
X
<( p /k)1/p = .
j=1

Thus, the countable set B is dense in (c00 , k · kp ) for each 1 ≤ p ≤ ∞ and hence c00
is separable.
e) Let 1 ≤ p < ∞. We prove that B is dense in (lp , k · kp ).
Let  > 0. Because we have seen that c00 is dense in (lp , k · kp ), we find for each x ∈ lp
an x0 ∈ c00 such that kx − x0 k < /2. Since B is dense in (c00 , k · kp ) we find x00 ∈ B
such that kx0 − x00 kp < /2. But then, kx − x00 kp < /2 + /2 = .

3. Banachs’s Fix-Point Theorem

Let X be a Banach space. A mapping φ : X → X is called a strict contraction if there is
a c ∈ (0, 1) such that kφ(x) − φ(y)k ≤ ckx − yk for all x, y ∈ X.

a) Prove that for a strict contraction φ there exists a unique fix-point, i.e. there is
exactly one point x∗ ∈ X such that φ(x∗ ) = x∗ .
b) Let X = (C([0, 1]), k · k∞ ) and let A := {f ∈ X : f (1) = 1}. Consider the map
M : A → A defined by M f (x) = xf (x). Prove that kM f − M gk∞ < kf − gk∞ for
all f, g ∈ A. Does M have a fix-point?
 Lösung:

a) Let x ∈ X and define a sequence (xn ) iteratively by x1 = x, xn+1 = φ(xn ) for n ∈ N.

We first prove that (xn ) is a Cauchy sequence. Let n, k ∈ N, n ≥ 2. Then,

kxn+k − xn k = kφ(xn+k ) − φ(xn )k ≤ ckxn+k−1 − xn−1 k

≤ ... ≤ cn−1 kxk+1 − x1 k ≤ cn−1 (kxk+1 − xk k + ... + kx2 − x1 k)
k−1
n−1
X cn−1
≤c cj kx2 − x1 k ≤ kx2 − x1 k.
1−c
j=0

n−1
For  > 0 there is N ∈ N such that c1−c kx2 − x1 k <  for n ≥ N and we find
kxn − xm k <  for n, m ≥ N . Thus, (xn ) is a Cauchy sequence and converges to some
x∗ ∈ X by completeness of X. We then have

kφ(x∗) − x ∗ k = lim kφ(x∗) − xn k = lim kφ(x∗) − φ(xn−1 )k

n→∞ n→∞
≤ c lim kφ(x∗) − xn−1 k = kφ(x∗) − x ∗ k.
n→∞

Since c < 1 we conclude kφ(x∗) − x ∗ k = 0, i.e. φ(x∗ ) = x∗ .

To see that x∗ is unique, assume there is y ∈ X with φ(y) = y. Then,

kx∗ − yk = kφ(x∗ ) − φ(y ∗ )k ≤ ckx∗ − yk.

Since c < 1 we conclude that y = x∗ .

b) Let f, g ∈ A with f 6= g and define a function h by h(x) := f (x) − g(x). Note that
x 7→ xh(x) defines a continuous function which is zero for x = 0, 1. Then, there is
x0 ∈ (0, 1) such that x0 h(x0 ) = maxx∈[0,1] xh(x). Then, we find

kM f − M gk∞ = max kxf (x) − xg(x)k = max kh(x)k

x∈[0,1] x∈[0,1]

= x0 h(x0 ) < h(x0 ) ≤ max kh(x)k∞ = kf − gk∞ .

x∈[0,1]

M has no fix-point. Because assuming there is f ∈ C([0, 1]) with M f = f , we find

xf (x) = f (x). From this we conclude that f (x) = 0 for x ∈ (a, b). But f (1) = 1 and
we conclude that f is not continuous on [0, 1], a contradiction.

Hausaufgaben

4.1. Dual Space

We consider the space c0 := {(xn )n∈N : xn ∈ C, limn→∞ xn = 0} equipped with the
supremum norm k · k∞ .
To show that (c0 )∗ ' l1 we construct an isometric isomorphism T : l1 → (c0 )∗ .

a) For x := (xn )n∈N ∈ c0 and y := (yn )n∈N ∈ l1 define T y(x) := ∞

P
n=1 xn yn .
Prove that T is continuous linear operator with kT yk(c0 )∗ ≤ kyk1 for all y ∈ l1 .
b) For a := (an )n∈N ∈ l1 we define a sequence (wn )n∈N with wn ∈ c0 via wk :=
(φ(a1 ), φ(a2 ), ..., φ(ak ), 0, ...), where φ(z) := z/|z| for z 6= 0 and φ(0) = 1. Consi-
der T a(wn ) for n → ∞ to conclude that kT ak(c0 )∗ = kak1 .
c) Let f ∈ (c0 )∗ and define a sequence y by yn = f (en ) where en denotes the standard
unit vector in CN . Prove that (yn )n∈N ∈ l1 and that kyk1 ≤ kf k(c0 )∗ .
d) Prove that T y = f .
 Lösung:

a) Given a ∈ l1 we find for all sequences x ∈ c0

∞
X ∞
X ∞
X
| an xn | ≤ |an | |xn | ≤ kxk∞ |an |.
n=1 n=1 n=1

Thus T is well defined and it clearly is linear. The above equation also shows that
kT ak(c0 )∗ ≤ kak1 and hence T is continuous.
b) By construction, kwN k∞ = 1 for all N ∈ N. Now, for a ∈ l1 we get
N N
N →∞
X X
T a(wN ) = an φ(an ) = |an | → kak1 .
n=1 n=1

From this we get kT ak(c0 )∗ = supkxk∞ =1 |T a(x)| ≥ kak1 . Together with a) we con-
clude that kT ak(c0 )∗ = kak1 for all a ∈ l1 and hence T is an isometry.
c) Let f ∈ (c0 )∗ and let yn := f (en ) for n ∈ N. Then, for all m ∈ N we get
m
X m
X m
X
|yn | = yn φ(yn ) = f (en )φ(yn )
n=1 n=1 n=1
m
X m
X
= |f ( φ(yn )en )| ≤ kf k(c0 )∗ k φ(yn )en k∞ = kf k(c0 )∗ .
n=1 n=1

We conclude that y ∈ l1 and that kyk1 ≤ kf k(c0 )∗ .

d) For x ∈ c0 we get
∞
X ∞
X ∞
X
T y(x) = yn x n = f (en )xn = f ( xn en ) = f (x).
n=1 n=1 n=1

Here we used the continuity of f .

4.2. Locally compact spaces

A topological space X is called locally compact if every point x ∈ X has a compact
neighbourhood.
Prove that a Banach space is locally compact if and only if it is finite dimensional.
Lösung:

In lecture we have seen that a normed space (X, k · k) is finite dimensional iff B 1 (0) is
compact.
Let us first prove that this implies that B r (x) is compact for all x ∈ X and all ∞ > r > 0.
For y ∈ X and s > 0 define the map ms,y : X → X, z 7→ s(z − y) − y and note that it is
a homeomorphism: ms,y is clearly injective, it is continuous because the pre-image of an
open ball is open and it is also open, because it maps open balls to open balls. Now let
{Ui }i∈I be an open cover of B r (x). Then {m−x,1/r (Ui )} is an open cover of B 1 (0). Thus
there is a finite sub-cover and the pre-image of this finite subcover gives a finite subcover
for B r (x).
Let (X, k · k) be a normed space and assume that it is locally compact. Let x ∈ M and let
V ⊂ X be a compact subset containing x. Then there is an  > 0 such that B2 (x) ⊂ V .
But then B  (x) ⊂ V is compact as a closed set of a compact set. But by the above, this
implies that B 1 (0) is compact and thus X is finite dimensional.
Conversely suppose (X, k · k) is a finite dimensional normed space and let x ∈ X. Then,
B 1 (x) is compact since it is closed and bounded.
4.3. Closed and Bounded Sets
Let (X, k · k) be a Banach space, C ⊂ X be a closed linear subspace and F ⊂ X be a
finite dimensional linear subspace.

a) Prove that C + F := {x + y : x ∈ C, y ∈ F } is a closed linear subspace.

b) Show that in general C + F is not closed for a closed linear subspace C ⊂ X, if we
require F ⊂ X to be a closed linear subspace.
Hint: Consider the linear subspaces X1 := {(xn )n∈N ∈ l2 : xn = 0 for n odd} and
X2 := {(xn )n∈N ∈ l2 : x2n = nx2n−1 for n ≥ 1}. Show that X1 + X2 is dense in l2
but that the sequence (1, 0, 1/2, 0, 1/3, 0, ...) is not in X1 + X2 .

Lösung:

a) Let X/C be the quotient space and let q : X → X/C be the quotient projection. We
have seen in the lecture that kq(x)kX/C := inf{kx − yk : y ∈ C} defines a norm on
X/C if C is closed. The quotient projection q is continuous since kq(x)kX/C ≤ kxk.
By linearity of q, π(F ) is a finite dimensional subspace of X/C and thus in particular
closed. Finally C + F = π −1 (F ) is closed by continuity of π.
b) Let c00 be the set of finitely supported sequences. Observe that (an )n∈N ∈ c00 can be
uniquely written as a sum of elements of X1 and X2 :

(x1 , x2 , x3 , x4 , .., xk , 0, ...)

k k
= (0, x2 − x1 , 0, x4 − 2x3 , ...0, xk − xk−1 , 0, ...) + (x1 , x1 , x3 , 2x3 , ..., xk−1 , xk−1 , 0, ...)
2 2
for k even and similarly for k odd. Thus c00 ⊂ X1 +X2 and we conclude that X1 +X2
is dense in l2 . Now consider the sequence (xn )n∈N ∈ l2 with xn = 1/n for n odd and
xn = 0 for n even. Then,

(1, 0, 1/2, 0, 1/3, 0, ...) = (0, −1, 0, −1, 0, −1, ...) + (1, 1, 1/2, 1, 1/3, 1, ...)

and hence (xn )n∈N ∈

/ X1 + X2 .