You are on page 1of 8

D-MATH Functional Analysis I HS 2012

Sample Solutions for Exercise Sheet 2

1. (a) Let X be a Banach space and Y a normed space. Let T : X → Y be a linear continuous
map which is not surjective. Show that T (X) ⊂ Y is a subset of the first Baire category.

Note: An earlier version of this exercise also required Y to be complete. This is not necessary;
see below and compare the updated solution for Exercise 1 on Exercise Sheet 1.

Replace `1 by X, C([0, 1]) by Y and f by T in the (sample) proof of Exercise 1 on Exercise


Sheet 1. Then note that this proof indeed used nothing else than

• the existence of a norm on the R-vector spaces `1 and C([0, 1]), and
• the completeness of `1 , and
• the linearity and continuity of f ,

in order to prove that the assumption “f (`1 ) is a subset of the second category in C([0, 1])”
implies “f (`1 ) = C([0, 1])”, i.e. f is surjective.
The assertion now follows by contraposition.

(b) Let 1 ≤ p < ∞. Let `p be the space of all real sequences (an ) which satisfy the inequality


!1/p
X
k(an )kp := |an |p < ∞.
n=1

Show that `p is dense in c0 (= the space of all real convergent sequences with limit 0, endowed
with the supremum norm) and a subset of the first Baire category.

The definition of the two spaces clearly implies that `p is a proper subset of c0 (e.g. for any
1/p 1 1/p 1 1/p
p ≥ 1, the sequence (1, 12 , 3 , 4 , . . .) is in c0 but not in `p ). Thus, the inclusion
p
function i : ` → c0 , a 7→ a is not well-defined and surjective; one also easily checks that it
is both continuous and linear. Thus, by part (a), the set `p = f (`p ) is a subset of the first Baire
category in c0 .
Let c00 be the set of all real sequences with only a finite number of non-zero terms. For all
sequences a = (an ) ∈ c0 and all positive integers k, we define sequences a(k) ∈ c00 by setting
(k) (k)
an = an if n ≤ k and an = 0 if n > k. Then ka − a(k) k∞ = supn>k |an | → 0 as k → ∞.
Thus c00 is dense in c0 . Since we also have c00 ⊂ `p ⊂ c0 , it follows that `p is dense in c0 .

(c) Let 1 ≤ p < q < ∞. Show that `p is dense in `q and a subset of the first Baire category.

We start by proving
k(an )kq ≤ k(an )kp (1)

1
for all (an ) ∈ `p and p < q. Let a = (an ) ∈ `p with kakp = 1. Then |an | ≤ 1 for all n ∈ N.
Since p < q, this yields |an |q ≤ |an |p for all n ∈ . Thus N
X X
kakqq = |an |q ≤ |an |p = kakpp = 1 = kakqp .
n≥1 n≥1

Thus (1) holds for all a ∈ `p with kakp = 1.


Now let b ∈ `p r {0} and set a := b/kbkp . Then kakq ≤ kakp = 1, as we have just seen. By
the homogenity property of the norms, we get

kbkq = kkbkp · akq ≤ kkbkp · akp = kbkp .

Let i : `p → `q , a 7→ a be again the inclusion function; it is linear and, by (1), well-defined.


Inequality (1) also implies that i is continuous. However, i is not surjective: We have

X
a := (n−δ ) ∈ `p ⇔ n−δp < ∞ ⇔ δp > 1.
n=1

Thus the sequence a is in `q but not in `p if 1q < δ < p1 . Part (a) now implies that `p is a subset
of the first Baire category in `q .
Like in (b), one easily shows that c00 is dense in `q . Since c00 ⊂ `p ⊂ `q , it follows that `p is
dense in `q .

2. Let p ≥ 1. Let Lp ([0, 1]) be the space of Lebesgue measurable functions f : [0, 1] → R with
Z 1 1/p
kf kp := |f (x)|p dx < ∞.
0

Show that L2 ([0, 1]) ⊂ L1 ([0, 1]) (endowed with the norm k · k1 ) is a subset of the first Baire
category in each of the following two ways:

(a) Use Exercise 1(a).

The Cauchy-Schwarz inequality implies

kf k1 = kf · 1k1 ≤ kf k2 k1k2 = kf k2 (2)

for all f ∈ L2 ([0, 1]); here 1 denotes the function which equals 1 on all of [0, 1]. Thus
N
L2 ([0, 1]) ⊂ L1 ([0, 1]). For each n ∈ , let χn be the characteristic function for the open
1
interval ( n+1 , n1 ). Then

R,
X
fδ : [0, 1] → x 7→ nδ χn (x)
n=1

is in Lp ([0, 1]) if and only if



!p ∞
1
nδp
Z X X
nδ χn (x) dx = <∞
0 n(n + 1)
n=1 n=1

2
which holds if and only if δ < p1 . Thus f3/4 ∈ L1 ([0, 1]) r L2 ([0, 1]), hence L2 ([0, 1]) is a
proper subset of L1 ([0, 1]).
The inclusion function i : L2 ([0, 1]) → L1 ([0, 1]) is therefore well-defined and continuous
(by inequality (2)) and not surjective. Therefore, by exercise 1(a), L2 ([0, 1]) is a subset of the
first category in L1 ([0, 1]).

(b) Show that the set


1
 Z 
1 2

B= f ∈ L ([0, 1]) |f (x)| dx ≤ 1
0

is closed in (L1 ([0, 1]), k · k1 ) with empty interior.

Note that [
L2 ([0, 1]) = n·B
n≥1

and recall that a closed set with empty interior is nowhere dense. In particular, the assertion of
(b) implies that L2 ([0, 1]) ⊂ L1 ([0, 1]) is a subset of the first category.
B is closed: Let (fn ) be sequence of functions in B which converges to some f ∈ L1 ([0, 1])
with respect to the metric induced by k · k1 , i.e.

kfn − f k1 → 0 as n → ∞.

Thus, by the relation between convergence in measure and pointwise convergence (as proved in
basic measure theory or analysis courses), there exists a subsequence (fnk ) such that fnk (x) →
f (x) as k → ∞ almost everywhere on [0, 1]. This yields |fnk (x)|2 → |f (x)|2 as k → ∞
almost everywhere on [0, 1]. Since fnk ∈ B, we have kfnk k2 ≤ 1 for all k. By Fatou’s Lemma
(the version for pointwise convergence) we therefore get
Z 1 Z 1
2
kf k2 = |f (x)| dx ≤ lim inf |fnk (x)|2 dx ≤ 1.
0 k→∞ 0

Thus, f ∈ B, i.e. B is closed.


The interior of B is empty: Assume that there exists f0 ∈ B and r > 0 such that

BL1 (f0 , r) = {g : kf0 − gk1 < r} ⊂ B.

For any f ∈ L1 ([0, 1]) we set


r f
gf = f0 + · .
2 kf k1
Then gf ∈ BL1 (f0 , r) ⊂ B. In particular, we have gf ∈ L2 ([0, 1]). This yields
2
f= · kf k1 · (gf − f0 ) ∈ L2 ([0, 1]),
r
in particular L2 ([0, 1]) = L1 ([0, 1]) which is a contradiction to what we have proved in (a).

3. Let X be a vector space and p : X → [0, ∞) a function satisfying:

3
(i) p(x) = 0 if and only if x = 0, and
(ii) p(λx) = |λ| p(x) for all x ∈ X and λ ∈ R.

Show that p is a norm if and only if the closed unit ball BX = {x ∈ X | p(x) ≤ 1} is convex.

Let p be a norm. For all x, y ∈ BX and all t ∈ [0, 1], we then have

p(tx + (1 − t)y) ≤ p(tx) + p((1 − t)y) = t · p(x) + (1 − t) · p(y) ≤ t + (1 − t) · 1 ≤ 1

where the first inequality follows from the triangle inequality, the equality by homogenity (note
that 1 − t ≥ 0) and the second inequality by the definition of BX . Thus, tx + (1 − t)y ∈ BX
for all t ∈ [0, 1] and therefore, by definition, BX is convex.

For the other direction, let BX be a convex set. The map p is positive homogeneous and positive
definite by the properties given in the assumptions of the exercise. It remains to show that p is
subadditive. If x = 0 and y ∈ X, then

p(x + y) = p(y) = 0 + p(y) = p(x) + p(y)

by property (i). Hence the triangle inequality holds in this case (and, of course, also in the case
y = 0, x ∈ X).
Let x, y ∈ X r {0}. Then p(x) + p(y) > 0 by property (i) and the non-negativity of p. The
triangle inequality is therefore equivalent to the inequality
p(x + y)
≤ 1. (3)
p(x) + p(y)
x y
Set a = p(x) and b = p(y) which are well-defined by (i) and the assumption on x and y. Note
that  
x (ii) 1
p(a) = p = · p(x) = 1
p(x) p(x)
and similarly p(b) = 1; thus a, b ∈ BX . Then
   
p(x + y) (ii) x+y x y
= p =p +
p(x) + p(y) p(x) + p(y) p(x) + p(y) p(x) + p(y)
 
p(x) x p(y) y
=p · + ·
p(x) + p(y) p(x) p(x) + p(y) p(y)
   
p(x) p(x)
=p ·a+ 1− ·b ≤1
p(x) + p(y) p(x) + p(y)
p(x)
where the last inequality holds by the convexity of BX (note that p(x)+p(y) ∈ [0, 1] by the
positivity of p on X r {0}). This proves inequality (3) and therefore the subadditivity of p.

4. Consider the following two norms on C([0, 1]) given by


Z 1
kf k∞ := sup |f (x)|, kf k1 := |f (t)| dt.
x∈[0,1] 0

4
For positive real numbers α and β we define

p(f ) := max(αkf k1 , kf k∞ ), q(f ) := min(βkf k1 , kf k∞ ).

(a) Find necessary and sufficient conditions on α such that p is a norm.

The function p clearly satisfies the properties that we assumed for the function p in Exercise 3.
Set
B = {f : p(f ) ≤ 1}, B1 = {f : kf k1 ≤ 1}, B∞ = {f : kf k∞ ≤ 1}.
Then f ∈ B if and only if αf ∈ B1 and f ∈ B∞ . By Exercise 3, we know that B1 and B∞ are
convex because k · k1 and k · k∞ are norms. Thus, for all f, g ∈ B and all t ∈ [0, 1], we have
1
(tf + (1 − t)g) ∈ B1 and tf + (1 − t)g ∈ B∞ .
α
Hence, tf + (1 − t)g ∈ B for all f, g ∈ B and all t ∈ [0, 1]. Again by Exercise 3, this implies
that p is a norm for all α > 0.

(b) Find necessary and sufficient conditions on β such that q is a norm.

Claim: q is a norm if and only if 0 < β ≤ 1.


Proof: Let 0 < β ≤ 1. Note that we have
Z 1
βkf k1 ≤ kf k1 = |f (t)| dt ≤ sup |f (x)| = kf k∞
0 x∈[0,1]

for all f ∈ C([0, 1]). Thus q = βk · k1 which is clearly a norm.


Let β > 1 and ε > 0. Let χ[x,y] denote the characteristic function for each interval [x, y] ⊂
[0, 1]. Set
x
f1 (x) = χ[0,1] , f2,ε (x) = (β − 1)(χ[0,ε/2] + 2(1 − )χ )
ε (ε/2,ε]
for all x ∈ [0, 1]. Then
1
3β(β − 1)
Z
ε ε ε
βkf2,ε (x)k1 = β |f2,ε (x)| dx = β(β − 1) · ( + 2 · − ε + ) = · ε.
0 2 2 4 4
Thus for every δ > 0, there exists ε > 0 such that q(f2,ε ) < δ. Now note that
 
3β(β − 1)
β = βkf1 k1 ≤ β 1 + · ε = βkf1 + f2,ε k1 ;
4
also note that
kf1 + f2,ε k∞ = β
for all ε > 0; hence q(f1 + f2,ε ) = β. Set δ := β − 1 > 0. Thus, there exists ε > 0 such that

q(f1 ) + q(f2,ε ) = 1 + q(f2,ε ) < 1 + δ = β = q(f1 + f2,ε )

which contradicts the triangle inequality.

Remark: There is nothing magical about the choice of f2,ε . The proof shows that every other
continuous function f2,ε with:

5
• minx∈[0,1] f2,ε (x) ≥ 0 and maxx∈[0,1] f2,ε (x) = β − 1 and
• kf2,ε k1 → 0 as ε → 0

would work.

5. Let  
n !
R
 
A = (xk )k∈N ⊂
X
x k converges
n∈N
 
k=1
and set n
X
kxk = sup xk

n∈N
k=1

for all x = (xk )k∈N ∈ A.

(a) Show that (A, k · k) is a Banach space.

A is a linear space: Let x = (xn ) and y = (yn ) be sequences in A. Then x + y = (xn + yn )


and the sequence
n n n
! !
X X X
(xk + yk ) = (xk )) + ( (yk )
k=1 n∈ N k=1 k=1 n∈ N
converges, i.e. x + y ∈ A. Moreover, the sequence αx lies obviously in A for all α ∈ R and
all x ∈ A.
A is a normed space:

• For every sequence x ∈ A, the sequence ( nk=1 xk )n is bounded and therefore kxk < ∞;
P
R
hence, the map k · k : A → is well-defined.
• k · k is clearly a non-negative map.
• If
Pkxk
n−1
= 0, then nk=1 xk = 0 for all n ∈
P
N; thus x1 = 0 and xn =
Pn
k=1 xk −
k=1 xk = 0 for all n ≥ 2; hence x = 0.
• For all x, y ∈ A and all α ∈ R, we also have
n n
X X
kαxk = sup αxk = sup |α| xk = |α|kxk

n∈N
k=1
n∈N
k=1

and

Xn Xn Xn
kx + yk = sup xk + yk ≤ sup xk + sup yk = kxk + kyk.

n∈N
k=1
n∈N
k=1 n∈N
k=1

6
This proves that k · k is indeed a norm on A.
A is complete: Let x(n) be a Cauchy sequence in A with respect to the norm k · k. Fix ε > 0;
N
then there exists Nε ∈ such that kx(n) − x(m) k < ε for all n, m ≥ Nε . Hence

X k X k−1
(n) (m) (n) (m) (n) (m)
|xk − xk | ≤ xi − xi + xi − xi ≤ 2kx(n) − x(m) k < 2ε


i=1 i=1

N N (n)
for all k ∈ . Therefore, for every k ∈ , the sequence (xk )n is a Cauchy sequence in . R
R (n)
Since is complete, there exists a sequence x = (xk ) with xk → xk as n → ∞ for all k.
Pj (n) (m)
For all n, m ≥ Nε , we also have supj | k=1 xk −xk | < ε. Passing to the limit as n → ∞,
Pj (m)
N
we get | k=1 xk − xk | ≤ ε for all m ≥ Nε and all j ∈ . Thus, kx − x(m) k ≤ ε for all
m ≥ Nε , so x(m) → x as m → ∞.
Finally, wePhave to show that x ∈ A. We use the Cauchy convergence P test which states that

the series k=1 xk converges if and only if the sequence of partial sums jk=1 xk is a Cauchy
N
sequence. For every C, D ∈ with D ≥ C, we have
D C
D D D
X X X X
(m) (m)
X
xk − xk = xk = xk − xk + xk . (4)



k=1 k=1 k=C+1 k=C+1 k=C+1

(m)
Since we know that ∞
P
k=1 (xk − xk ) converges, the Cauchy convergence test implies that
N
there exists Kε ∈ such that the first term on the right-hand side of (4) is < ε for all C, D ≥
Kε . Analogously, the second term on the right-hand side of (4) is < ε for all C, D ≥ Lε for
N
some Lε ∈ , because ∞
P
k=1 xk
(m)
converges since x(m) ∈ A. Thus, for every ε > 0 we can
take E = max(Kε , Lε ) and have

X D X C
xk − xk < 2ε



k=1 k=1
Pj
for all C, D ≥ E, i.e. the sequence of partial sums k=1 xk is indeed a Cauchy sequence.

(b) Show that (`1 , k · k) is not a Banach space.


k
Consider the alternating harmonic sequence a = (ak ) with ak = (−1) . By the alternating
P∞ (−1)k k
series test, we know that the alternating harmonic series k=1 k converges (and is equal
(j)
to log 2). Thus, a ∈ A. Define the sequence b(j) = (bk )k by setting
(
(j) ak , if k ≤ j,
bk =
0, otherwise .
P k

Then ka − b(j) k = supn∈N nk=j+1 (−1) k tends to 0 as j → ∞ (note: this is not true if

k · k is replaced by k · k1 ); in particular, b(j) is a Cauchy sequence with respect to k · k. For


N
all j ∈ , we have b(j) ∈ `1 (asPa sequence with only finitely many non-zero elements), but
/ `1 (because n≥1 |an | = n≥1 n1 diverges). Hence `1 is not complete with respect to
P
a ∈
k · k.

7
(c) Show that `1 is a dense subset of A.

Let x = (xn ) ∈ A.PFor all ε > 0 there exists Nε ∈ N such that for all p ≥ 1 and all
n+p ε
n ≥ Nε , we have | i=n+1 xi | < 2 . Choose Mε ∈ N such that 21n < 2ε for all n ≥ Mε .
Set K = max(Nε , M −(K+1) , 2−(K+2) , . . .). Then z ∈ `1 and
ε ) and z = (x1 , . . . , xK , 2
r
kz − xk = supr>K i=K+1 (2−i − xi ) . Moreover, by the definition of K and Mε , we also
P
have r
X Xr X r
(2−i − xi ) ≤ 2−i + xi ≤ 2−K + ε/2 < ε.



i=K+1 i=K+1 i=K+1

Thus, kz − xk < ε and therefore `1 is dense in A.


Note: This argument proves the stronger statement that `1 r c00 , i.e. the set of sequences in
`1 with infinite support, is dense in A. For the exercise at hand, it would suffice to take an
appropriate sequence z ∈ c00 (cf. Exercise 1(b)), show that c00 is dense in A and then use the
inclusions c00 ⊂ `1 ⊂ A.

You might also like