Applied Functional Analysis, Solutions 2

15.11.2007

Exercise 1 a) We have fn := en − 2en+2 , n ∈ N, and S := {fn | n ∈ N}. To

compute S ⊥ , take x ∈ `2 arbitrary. Then x ⊥ S iff x ⊥ fn for all n ∈ N iff

xn − 2xn+2 = hx, en i − 2 hx, en+2 i = hx, fn i = 0

for all n ∈ N. This shows that x ∈ S ⊥ iff

1
xn+2 = xn (n ∈ N)
2
Hence the vectors
1 1
h1 := (1, 0, , 0, , 0, . . . )
2 4
1 1
h2 := (0, 1, 0, , 0, , 0, . . . )
2 4

are certainly in S ⊥ . On the other hand, given h ∈ S ⊥ the vector

y := x − x1 h1 − x2 h2

satisfies y1 = y2 = 0 and yn+2 = 12 yn for all n ≥ 3. Hence y = 0 and so

x = x1 h1 + x2 h2 . Consequently, S ⊥ = span{h1 , h2 }. But obviously h1 ⊥ h2 ,
so {h1 / kh1 k , h2 / kh2 k} is an orthonormal system spanning F = S ⊥ .
b) For a general x ∈ `2 we have
h1 h2 hx, h1 i hx, h2 i
PF x = hx, h1 / kh1 ki + hx, h2 / kh2 ki = 2 h1 + h2 .
kh1 k kh2 k kh1 k kh2 k2
We need to compute kh1 k2 and kh2 k2 . Since h2 has the same “coordinates”
as h1 apart from an additional 0,
∞ ∞  n
2 2
X
−n 2
X 1 1 4
kh1 k = kh2 k = (2 ) = = = .
n=0
4
n=0
1 − 14 3

Now specializing x = e1 + e2 we obtain hx, h1 i = 1 = hx, h2 i and hence

1 1 3 3 1 1 1 1
PF x = h1 + h2 = (h1 + h2 ) = (1, 1, , , , , . . . ).
4/3 4/3 4 4 2 2 4 4
2 Applied Functional Analysis Solutions 2

Exercise 2 Let f ∈ Lp (a, b), and let q the conjugated exponent to p, i.e. it
holds 1/p + 1/q = 1. By Hölder’s inequality, for x ≤ y
Z y  Z y
 
|(Jf )(y) − (Jf )(x)| = 
 f (s) ds ≤ |f (s)| ds
x x
Z y 1/p Z y 1/q
≤ |f (s)|p ds 1(s)q ds
x x
1/q 1− p1
≤ kf kp |y − x| = kf kp |y − x| .

This shows that Jf is Hölder continuous with exponent α = 1 − 1/p, and

kf k(α) ≤ kf kp (f ∈ Lp (a, b)).

1−1/p
Since (Jf )(a) = 0 is obvious, we proved f ∈ C0 [a, b] as claimed.

Exercise 3 a) If f = c is constant, then by the fundamental theorem of

calculus Z b Z b
0
cϕ ds = c(ϕ(b) − ϕ(a)) = 0 = − 0 · ϕ(s) ds
a a
for every ϕ ∈ C01 [a, b].
Hence by definition, 0 is a weak derivative of f . (Since
we proved in the lectures that weak derivatives are unique, we may even say:
0 is the weak derivative of f .)
Alternative: As f is constant, f ∈ C 1 [a, b] with usual derivative f 0 = 0. In
Lecture 6 (second remark after Definition 11) we proved that in this case the
usual derivative is a weak derivative. Hence 0 is a (the) weak derivative of a
constant function.
Remark: A common mistake here was to start with something like: “Let f
be constant and g its weak derivative.” and in the end it is shown that g = 0.
But: in general a weak derivative does not exist! And so one has only proved:
IF a weak derivative of a constant function f exists, then it is zero. And this
is not what was asked for.

b) Suppose that f ∈ C[a, b] can be written as f = ϕ0 + c1 with c ∈ R and

ϕ ∈ C01 [a, b]. Then

hf, 1i = ϕ0 , 1 + c h1, 1i = c(b − a).



Rb
This shows that c = (b−a)−1 a f (s) ds is uniquely determined by f . To prove
the claim of the exercise, let f ∈ C[a, b] be arbitrary, define c := (b−a)−1 hf, 1i
and g := f − c1. Then ϕ defined by
Z x
ϕ(x) := g(s) ds (x ∈ [a, b])
a
Solutions 2 Applied Functional Analysis 3

clearly is in C 1 [a, b], ϕ0 = g, and ϕ(a) = 0. But also

Z b
ϕ(b) = g(s) ds = hg, 1i = hf − c1, 1i = hf, 1i − c h1, 1i = hf, 1i − hf, 1i = 0
a

and hence ϕ ∈ C01 [a, b].

Remark: A common mistake here was that there was no clear distinction
between the first part of the answer (getting an idea how ϕ and c should be)
and then proving the claim. Many of you actually proved: IF f can be written
in the form f = ϕ0 + c with ϕ, c as required, THEN c has the form . . . and ϕ
has the form . . . . But strictly speaking, this is not what was asked for.
And attention: if you use an implication sign as

⇒ ϕ = ...

this has to be read as a statement about an object ϕ, that was already intro-
duced. If you actually mean “Therefore, we can chose ϕ := ... then you have
to write that, and not use the form above!

c) We have seen in a) that ϕ0 ⊥ 1 for every ϕ ∈ C01 [a, b]. This proves the
inclusion “⊃”. To prove the reverse inclusion take f ∈ H = L2 (a, b) such that
f ⊥ ϕ0 for all ϕ ∈ C01 [a, b]. Define e to be the constant function e := (b−a)−1 1.
Thus h1, ei = 1. Now.
f − hf, 1i e ⊥ ϕ0
for all ϕ ∈ C01 [a, b], but also

f − hf, 1i e ⊥ 1.

Hence by b),
f − hf, 1i e ⊥ C[a, b].
But C[a, b] is dense in L2 (a, b) and so

f − hf, 1i e ⊥ L2 (a, b)

since the scalar product is continuous (Lecture 6, Corrolary 7 or LN 7.2.4).

This implies that f − hf, 1i e = 0, i.e. f = hf, 1i e = hf,1i
b−a 1 almost everywhere.

Here is a different way to prove the second half of c). Define S := {ϕ0 | ϕ ∈
C01 [a, b]} and again e := (b − a)−1 1, and fix f ∈ S ⊥ . Since C[a, b] is dense in
L2 [a, b], we find a sequence (fn )n ⊂ C[a, b] such that fn → f in the L2 -norm.
Now, by 3b) we can write fn = ϕ0n + cn , with certain ϕn ∈ C01 [a, b] and cn ∈ R.
Then
hfn , ei = ϕ0n , e + hcn 1, ei = 0 + cn h1, ei = cn .


4 Applied Functional Analysis Solutions 2

Define c := hf, ei, then limn cn = limn hfn , ei = hf, ei = c. Consequently,

ϕ0n = fn − cn 1 → f − c1

in L2 (a, b). This shows that f − c1 ∈ S. But f, 1 ⊥ S, and hence f, 1 ⊥ S

⊥
(Lemma 2, Lecture 7). So f − c1 ∈ S ∩ S = {0}. Hence f = c1 is constant.

d) Suppose that f ∈ H 1 (a, b) such that f 0 = 0. Then

f, ϕ0 = − f 0 , ϕ = 0




for all ϕ ∈ C01 [a, b], by the definition of a weak derivative. Now this is nothing
else than
f ⊥ {ϕ0 | ϕ ∈ C01 [a, b]}
which by c) implies that f is a constant.

Exercise 4 a) Cauchy-Schwarz yields

Z b Z t 2 Z b Z t 
kJf k22 2
 
=  f (s) ds dt ≤ |f (s)| ds (t − a) dt
 
a a a a
Z b
(b − a)2
≤ (t − a) dt kf k22 = kf k22 .
a 2

This yields kJf k2 ≤ b−a

√ kf k and hence that J : L2 (a, b) −→ L2 (a, b) is
2 2
continuous, (see Assignment 1, Exercise 1, or Lecture 9).
Remark: Other estimates are valid. For example
Z b Z t 2 Z b Z b 2
kJf k22
 
=  f (s) ds dt ≤ |f (s)| ds dt
 
a a a a
Z b
≤ kf k2 (b − a) dt = (b − a)2 kf k22
a

which leads to kJf k2 ≤ (b − a) kf k2 . Also, one may use Exercise 2, which

then amounts to similar estimates.

b) Let f ∈ C[a, b] and ϕ ∈ C01 [a, b]. Then

Z b Z t 
0
f (s) ds ϕ0 (t) dt


Jf, ϕ =
a a
Z b  Z a  Z b
= f (s) ds ϕ(b) − f (s) ds ϕ(a) − f (s)ϕ(s) ds
a a a
= − hf, ϕi
Solutions 2 Applied Functional Analysis 5

by ordinary integration by parts. Now let f ∈ L2 (a, b) be arbitrary and fix

ϕ ∈ C01 [a, b]. By the density of C[a, b] in L2 (a, b), there is a sequence (fn )n ⊂
C[a, b] such that kfn − f k2 → 0. Since J is continuous with respect to the
2-norms, kJfn − Jf k2 → 0 as well. Since the scalar product is continuous
(Lecture 6, Corrolary 7 or LN 7.2.4),

Jfn , ϕ0 → Jf, ϕ0



and hfn , ϕi → hf, ϕi

as n → ∞. By what we know already hJfn , ϕ0 i = − hfn , ϕi. It follows that

Jf, ϕ0 = − hf, ϕi



as claimed.
c) The statement in b) tells us that f is a weak derivative of Jf , for any
f ∈ L2 (a, b). I particular, Jf ∈ H 1 (a, b). Since 1 ∈ H 1 (a, b) anyway, we have
the inclusion “⊃”. Now let h ∈ H 1 (a, b) be arbitrary. Define f := h0 ∈ L2 (a, b).
Then Jf and h have the same weak derivative, i.e., the weak derivative of
h − Jf is 0. By Exercise 3 this implies that h − Jf = c1 for some c ∈ R. This
proves the other inclusion.
Remark. Sometimes the last argument was phrased in a somewhat different
way: take h ∈ H 1 (a, b) and define f = h0 and c := h(a). And then one tries
to show that h = Jf + c. The mistake here is that the expression “h(a)” is
not defined yet. Originally h is just a function from L2 (a, b) with some special
property (it has a weak derivative), and as such is just an equivalence class
of measurable functions. So you may change h on the null-set {a} as you like
without effectivley changing h as an element of L2 .
The point is, that after this exercise, we will know that there is a continuous
function h1 such that h = h1 almost everywhere. And then h(a) is defined by
using this (uniquely determined) continuous representative: h(a) := h1 (a).
To sum up: The upshot of Exercises 3 and 4 is to justify the formula
Z t
h(t) = h(a) + h0 (s) ds
a

for functions h ∈ H 1 (a, b), given that “h(a)” is meaningless in the first place.