Ben Carriel Math 6110 Final Exam December 12, 2012

Exercise 4.7.28:
(a) We can use the fact that L2 is self-dual so that Hölder’s inequality becomes
Z  Z 1/2 Z 1/2
2 2
 
 K(x, y)f (y)dy  ≤ |K(x, y)| dy |f (y)| dy
 
B B B

Applying the estimate |K(x, y)| ≤ A|x − y|α−d we see that

Z  Z 1/2 Z 1/2
2 2(α−d) 2
 
 K(x, y)f (y)dy  ≤ A |x − y| dy |f (y)| dy
 
B B B
 Z 1/2 Z 1/2
2 2(α−d) 2
≤ A 2 1dy |f (y)| dy
B B
Z 1/2
α−d 1/2 2
= A2 m(B) |f (y)| dy
B

where m(B) is the measure of the unit ball in Rd . Because f ∈ L2 (B) we have that
the right hand side above is bounded and hence, if we square the above inequality and
integrate (with respect to x) we get

kT f k2L2 (B) ≤ (A2α−d m(B)1/2 )2 kf k2L2 (B)

Equivalently,
kT f kL2 (B) ≤ A2α−d m(B)1/2 kf kL2 (B)
So we have that

kT k = inf{M | kT f kL1 ≤ M kf kL2 } ≤ A2α−d m(B)1/2

Which means that T is a bounded operator on L2 (B).

(b) Following the hint, define
(
K(x, y) |x − y| ≥ 1/n
Kn (x, y) =
0 otherwise

and Z
Tn (f )(x) = Kn (x, y)f (y)dy
B
We begin by verifying the following

Claim. Each of the operators Tn is a compact operator.

Proof. This is clear if we note that we just apply Exercise 4.7.26 to each of the Tn . this
exercise was proved on the homework and gives that each operator is compact.

We now want to show that Tn → T in the operator norm. Indeed,

Z Z
|Kn (x, y) − K(x, y)|dy ≤ A|x − y|α−d dy
B |x−y|≤1/n

Then if we set Z
1
Mn = dz
|z|≤1/n |z|d−α

1
Ben Carriel Math 6110 Final Exam December 12, 2012

Then 1/|z|d−α ∈ L1 (Rd ) because the exponent is less that d. Consequently, the
absolute continuity of the integral implies that Mn → 0 as n → ∞. Substituting this
in the above gives that
Z
|Kn (x, y) − K(x, y)|dy ≤ AMn
B

Which also goes to 0 as n → ∞ and so kTn − T k → 0. Because each Tn is compact

and Tn → T we can apply Proposition 6.1 to get that T is compact as well.
Exercise 6.7.8:
(a) We note that we can decompose Q1 into a disjoint union of exactly nd cubes of side
length 1/n. Each of these cubes is a translate of the cube Q1/n . So because the union
is disjoint we see that
d
n
X
µ(Q1 ) = µ(Q1/n ) = nd µ(Q1/n )
k=1

Dividing out nd gives that

m(Q1/n ) = n−d µ(Q1 )
And if we let µ(Q1 ) = c then we have that µ(Q1/n ) = cn−d .
(b) We want to see that µ  m. So we need to see that for every Lebesgue measurable
set E with measure 0, then µ(E) = 0. Indeed, choose a µ-measurable set E with
m(E) = 0. Then for any  > 0 we can find an open set O such that m(O − E) < .
Consequently, we see that m(O < . Then we can decompose O into a countable
union of almost disjoint cubes Qj with side length 1/n for some integer n (Theorem
1.1.4). Then by the previous part µ(Qj ) = cm(Qj ) and so
∞
X ∞
X
µ(O) = µ(Qj ) = cm(Qj ) = cm(O < c
j=1 j=1

Letting  → 0 shows that µ(E) = 0. Thus, µ  m and we can apply Theorem 6.4.3
to get that Z
µ(E) = f dm
E
We then immediately get that f is locally integrable because it must be integrable on
every µ-measurable set (Borel sets) and in particular, every ball.
(c) Let x be a point in the Lebesgue set of f and let Qn be a collection of half-open dyadic
rational cubes that contain x. Then the family Qn shrinks regularly to x because the
ratio of a cube Q of size length n to its circumscribing ball is given by
√
m(Q) nd nd 2
= d = −1/2 d =
m(BQ ) r vd m(B1 (0)) 2 n vd m(B1 (0)) vd m(B1 (0))

where B1 (0) is the unit ball about the origin and vd is a constant depending only on
d, and hence the ratio is constant. For each cube we have that
Z
1 1 cm(Qn )
f dm = · µ(Qn ) = =c
m(Qn ) Qn m(Qn ) m(Qn )

So f (x) = c at every Lebesgue point of f . We then note that because f is locally

integrable almost every point of Rd is in the Lebesgue set of f and so f (x) = c almost
everywhere.

2
Ben Carriel Math 6110 Final Exam December 12, 2012

Exercise 1.8.10:
(a) To see that Rd with Lebesgue measure is separable we set the collection of sets {Ek }∞
k=1
to be all finite unions of cubes with side length 1/n with rational endpoints for some
integer n. Then for each measurable set E in Rd and  > 0 we can choose a family of
cubes such that
∞
[ X∞
E⊂ Qj and m(Qj ) ≤ m(E) + /2
j=1 j=1

This choice can always be made because if E is measurable then we can find an open
set O ⊃ E such that m(O − E) <  and we can choose a set of cubes Qj such that
P ∞
j=1 m(Qj ) = m(O).

Since m(E) < ∞, the series converges and so we can find an N large enough that
P SN
j>N m(Qj ) < /2. Then if F = j=1 Qj we see

m(E4F ) = m(E − F ) + m(F − E)

   
[ [
≤ m Qj  + m  Qj − E 
j>N j≤N

X ∞
X
≤ m(Qj ) + m(Qj ) − m(E)
j>N j=1

≤

Letting  → 0 gives the result.

(b) We need to find the countable dense subset in Lp (X). Consider all functions of the
form r χEk (x) where r is a complex number with rational real and imaginary parts
and Ek is the family of measurable sets in X such that µ(E) < ∞ implies that
µ(E4Enk ) → 0 as k → ∞. I claim that the set of finite linear combinations of these
functions is dense in Lp (X).
Suppose that we have an f ∈ Lp (X) and for each n ≥ 1 the function
(
f (x) if x ∈ En and |f (x)| ≤ n
gn (x) =
0 otherwise

Then we have that |f − gn |p ≤ 2p |f |p so f − gn is in Lp (X). Furthermore, f ∈ Lp (X)

implies that f (x) ≤ ∞ almost everywhere and so gn (x) → f (x) almost everywhere.
We then apply the dominated convergence theorem to get that kf − gn kLp (X) → 0 as
n → ∞. Thus, for sufficiently large N we have kf − gN kLp (X) < /2. Let g = gN
and observe that g is a bounded function supported on a set of finite measure, so
g ∈ L1 (X). In particular, g is measurable and can therefore be approximated by a
sequence of increasing, simple functions ϕk % g such that each for large enough M
we have ϕM ≤ N and
p
Z
|ϕM − g|dµ ≤
X 4(2N )p−1
We then note that because ϕM = j=1 αj χAj where the αj ∈ C and Aj is measurable.
PD

3
Ben Carriel Math 6110 Final Exam December 12, 2012

So we can find a simple function ψ =

PD
j=1 rj
χEj such that
Z D Z
αj χAj (x) − rj χEj (x)dµ
X
|ϕ(x) − ψ(x)|dµ ≤
X j=1 X

D Z
αj χEj (x) − rj χEj (x) + αj χAj −Ej dµ
X
≤
j=1 X

D Z
(αj − rj ) χEj (x) + αj χAj −Ej dµ
X
≤
j=1 X

We then choose rj such that

p
|αj − rj | ≤ Pd
2p+1 (2N )p−1 j=1 µ(Ej )
and choose the Ej such that if α = maxj≤D αj then
p
µ(Aj − Ej ) ≤
2p+1 (2N )p−1 Dα
So that the above becomes
D Z D Z D Z
(αj − rj ) χEj (x) + αj χAj −Ej dµ ≤ χEj dµ + χAj −Ej dµ
X X X
|αj − rj | α
j=1 X j=1 X j=1 X

D d
p X X
≤ P D
µ(Ej ) + α µ(Aj − Ej )
2p+1 (2N )p−1 j=1 µ(Ej ) j=1 j=1
p p
≤ + Dα ·
2p+1 (2N )p−1 2p+1 (2N )p−1 Dα
p

= 2 · p+1
2 (2N )p−1
p
= p+1
2 (2N )p−1
Hence, we can approximate f as follows
kf − ψkLp (X) ≤ kf − gkLp (X) + kg − ψkLp (X)
We have already shown that with our choice of g the first term is bounded by /2. For
the second term we compute
Z
p
kg − ψkLp (X) = |g − ψ|p dµ
X
Z
= |g − ψ|p−1 |g − ψ|dµ
X
≤ 2p−1 (|g|p−1 − |ψ|p−1 )|g − ψ|dµ
Z
≤ 2(2N )p−1 |g − ψ|dµ
X
Z Z 
≤ 2(2N )p−1 |g − ϕ|dµ + |ϕ − ψ|dµ
 X p
X
p


≤ 2(2N )p−1 +
2p+1 (2N )p−1 2p+1 (2N )p−1
p

= p
2

4
Ben Carriel Math 6110 Final Exam December 12, 2012

Taking pth roots gives kg − ψkLp (X) < /2. Plugging this back into the triangle
inequality gives that
kf − ψkLp (X) ≤ 
So Lp (X) is separable.
PN
Exercise 1.8.16: We are given a finite set of functions fj ∈ Lpj (X), where j=1 1/pj = 1
whenever pj ≥ 1 for each j. We need to verify that
N
Y N
Y
k fj kL1 (X) ≤ kfj kLpj (X)
j=1 j=1

PN
It will actually we be easier to prove the more general case when j=1 1/pj = r for some r
and so
N
Y N
Y
k fj kLr (X) ≤ kfj kLpj (X)
j=1 j=1

This is a slight abuse of notation because it may be the case that r is not actually a norm.
In this case we still are referring to the quantity
Z 1/r
kf kLr (X) = |f |r dµ
X

We will proceed by induction on N . The case N = 1 is the familiar fact that

Z r Z
|f |r dµ
 
 f dµ ≤
 
X X

For the case N = 2 we apply the usual Hölder inequality to |f1 |r , g = |f2 |r and p1 /r, p2 /r
as the exponents. This gives that
Z
kf1 f2 krLr (X) = |f1 (x)f2 (x)|r dµ
X
≤ k|f1 |r kLp1 /r (X) · k|f2 |r kLp2 /r (X)
Z r/p1 Z r/p2
r(p1 /r) r(p2 /r)
= |f1 | dµ |f2 | dµ
X X
= kf1 krLp1 (X) · kf2 krLp2 (X)

So the result is established in the N = 2 case. We then proceed to the inductive step and
suppose that
n−1
Y n−1
Y
k fj kLr (X) ≤ kfj kLpj (X)
j=1 j=1

To show that the inequality is valid for n we simply compute

n
Y n−1
Y
k fj kL1 (X) = k( fj ) · |fn |k
j=1 j=1

Qn−1
We then apply the usual Hölder’s inequality with j=1 fj and fn noting that 1/pn +
Pn−1 P −1
n−1
j=1 1/pj = 1 so if we let r = j=1 1/pj then this becomes

n
Y n−1
Y
k fj kL1 (X) ≤ k fj kLr (X) · kf kLpn (X)
j=1 j=1

5
Ben Carriel Math 6110 Final Exam December 12, 2012

We then apply the n − 1 inequality to get the desired result.

Exercise 4.6.11:
Let M[a,b] denote the set of continuous functions that are not monotonic in [a, b] with
a, b ∈ Q. We need to show that each of these sets must be open in C([0, 1]). If we choose
an f ∈ M[a,b] then it measn that f is not monotonic in [a, b] and therefore we can find
points x0 < x1 < x2 such that f (x0 ) < f (x1 ) < f (x2 ). If we compute the distance between
f and some other function g we see that g is closer to f than the smaller of the numbers
f (y) − f (z) and f (y) − f (z). As a result we also see that g(x) < g(y) and g(z) < g(y) so
that g is not monotonic on M[a,b] either and so M[a,b] is open.
We have seen that M[a,b] is open and so it is everywhere dense.