(Contemporary Soviet Mathematics) v. M. Alekseev, V. M. Tikhomirov, S. v. Fomin (Auth.) - Optimal Control-Springer US (1987)
(Contemporary Soviet Mathematics) v. M. Alekseev, V. M. Tikhomirov, S. v. Fomin (Auth.) - Optimal Control-Springer US (1987)
(Contemporary Soviet Mathematics) v. M. Alekseev, V. M. Tikhomirov, S. v. Fomin (Auth.) - Optimal Control-Springer US (1987)
Contents
1 Chapter 1 3
1.1 Folland 1.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Folland 1.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Folland 1.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Boxes vs cylinder sets w.r.t. σ-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Folland 1.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Folland 1.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7 Folland 1.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.8 Folland 1.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.9 Folland 1.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.10 Folland 1.17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.11 Folland 1.18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.12 Folland 1.26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.13 Folland 1.28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.14 Folland 1.30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.15 Folland 1.31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.16 Folland 1.33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 Chapter 2 15
2.1 Folland 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Folland 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Folland 2.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Folland 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Folland 2.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Folland 2.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7 Folland 2.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.8 Folland 2.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.9 Folland 2.12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.10 Folland 2.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.11 Folland 2.14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.12 Folland 2.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.13 Folland 2.17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.14 Differentiable functions are Borel Measurable . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.15 Folland 2.20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1
2.16 Folland 2.21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.17 Folland 2.24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.18 Folland 2.34 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.18.1 Folland 2.33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.19 Folland 2.39 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.20 Folland 2.42 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.21 Folland 2.44: Lusin’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.22 Folland 2.46 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.23 Folland 2.48 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.24 Folland 2.49 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Chapter 3 30
3.1 Folland 3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Folland 3.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Folland 3.12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Folland 3.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Folland 3.17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Folland 3.20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Folland 3.21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Folland 3.24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.9 Folland 3.25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.10 Folland 3.26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Chapter 5 39
4.1 Folland 5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Folland 5.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Folland 5.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 Folland 5.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.5 Folland 5.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5 Chapter 6 44
5.1 Folland 6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Folland 6.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.3 Folland 6.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.4 Folland 6.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.5 Folland 6.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.6 Folland 6.14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2
1 Chapter 1
1.1 Folland 1.2
Prove the following Proposition:
Proposition. 1.1:
BR is generated by each of the following:
(a) the open intervals: E1 = {(a, b) | a < b},
(b) the closed intervals: E2 = {[a, b] | a < b},
Proof. Most of the proof is already completed by Folland. What was shown is that M(Ej ) ⊂ BR ∀j =
1, . . . , 8. To finish the proof and show BR = M(Ej ) ∀j, we can simply show that BR ⊂ M(Ej ) ∀j.
By invoking Lemma 1.1, if the family of open sets lie in M(Ej ), then it must be that BR ⊂ M(Ej ).
Furthermore, it is actually sufficient to only show that all the open intervals lie in M(Ej ) since every
open set in R is a countable union of open intervals. Thus, we complete our proof by directly showing
the following:
1. (a, b) ∈ E1 ⇒ (a, b) ∈ M(E2 ).
2. (a, b) = ∪∞
1 [a + n
−1
, b − n− 1] ∈ M(E2 )
1 (a, b − n 1] ∈ M(E3 )
3. (a, b) = ∪∞ −
4. (a, b) = ∪∞
1 [a + n
−1
, b) ∈ M(E4 )
c
5. (a, b) = (a, ∞) ∩ (−∞, b) = (a, ∞) ∩ [b, ∞)c = (a, ∞) ∩ ∩∞ 1 (b − n
−1
, ∞) ∈ M(E5 )
−1 c
6. (a, b) = (a, ∞) ∩ (−∞, b) = (−∞, a]c ∩ (−∞, b) = ∩∞ ) ∩ (−∞, b) ∈ M(E6 )
1 (−∞, a + n
3
Proof. The forward direction (σ-algebra ⇒ closed under countable increasing unions) is by the definition
of σ-algebra (closed under countable unions ). The backward direction (closed under countable increasing
unions ⇒ closed under countable increasing unions ⇒ σ-algebra) is slightly more involved:
j
1 ∈ A, then let us define Ej := ∪1 Fi . Since countable unions of countable unions is countable,
If {Fi }∞
and since {Ej }1 has the property of E1 ⊂ E2 ⊂ · · · , then we know that ∪∞ 1 Ej ∈ A. However, since
∞
If M(E) is the σ-algebra generated by E, then M(E) is the union of the σ-algebras generated by
Fα as Fα ranges over all countable subsets of E.
Proof. We use the notation Fα to denote a countable subset of E, and we let F := {Fα | α ∈ A} denote
the (likely uncountable) set of all countable subsets of E. Let us also define M̂ := ∪α∈A M(Fα ). We
proceed now by first showing that M̂ is indeed a σ-algebra by showing that M̂ is closed under countable
unions and compliments:
Suppose {Ei }∞ 1 ∈ M̂. Since M̂ is simply the union of a many σ-algebras, we know immediately that ∀Ei
∃ at least one Fi s.t. Ei ∈ M(Fi ). Since a countable union of countable elements is countable, if we define
H := ∪∞ 1 Fi where Ei ∈ M(Fi ), we know that H is also countable subset of E. We can now look at the
properties of the following σ-algebra: M(H).
(1) Since Fi ⊂ H ⊂ M(H) ⇒ M(Fi ) ⊂ M(H) (by Lemma 1.1), and since Ei ∈ M(Fi ), we can say that
1 ∈ M(H).
{Ei }∞
(2) Since H is a countable subset of E, we know that ∃β s.t. H = Fβ , and hence M(H) ⊂ M̂.
Therefore, since M(H) is by construction a σ-algebra and from (1) ({Ei }∞
1 ∈ M(H)) it ⇒ ∪1 Ei ∈ M(H),
∞
To now show M̂ is closed under compliments, suppose E ∈ M̂. By the same argument already used,
there must exist a countable subset Fα ⊂ E s.t. E ∈ M(Fα ), and obviously since M(Fα ) is a σ-algebra,
E c ∈ M(Fα ). Therefore, since M(Fα ) ⊂ M̂ ⇒ E c ∈ M̂. We have thus shown that M̂ is is closed under
countable unions and compliments, and hence a σ-algebra.
To neatly finish up our proof, let us first note that ∀α ∈ A, Fα ⊂ E ⇒ M(Fα ) ⊂ M(E), and thus we
can also say M̂ ⊂ M(E). To show the opposite relation, let ε ∈ E, then ε is trivially countable, so ∃β
s.t. ε = Fβ ⇒ ε ∈ M̂. Now since this is true ∀ε ∈ E, we can say that E ⊂ M̂, which therefore (again by
Lemma 1.1) ⇒ M(E) ⊂ M. By showing both opposite relations, we can thus conclude that M(E) = M̂.
4
1.4 Boxes vs cylinder sets w.r.t. σ-algebras
Exercise. 1.1:
Let A be an index set, {Xα }α∈A a family of non-empty sets and for each α ∈ A, Mα be a σ-algebra
on Xα . Consider the product space: Y
X= Xα
α∈A
Let M be the σ-algebra generated by the cylinder sets C := {πα−1 (Eα ) | Eα ∈ Mα , α ∈ A}, and
M be the one generated by boxes
∗
B
:= { α∈A Eα | Eα ∈ Mα }. Show that M ⊂ M∗ , but in
Q
general M 6= M∗
Hint 1: Proposition 1.3 implies that if A is countable then M = M∗ ; we should thus take A to be
not countable.)
Hint 2: You might find useful to first prove the following intermediate result. For any A0 ⊂ A, let
MA0 = M({π −1 (Eα ) | Eα ∈ Mα , α ∈ A0 }); let now
[
M̃ = MA0
A0 ⊂A countable
Then show that M = M̃. (Hint2 : show that M̃ is a σ-algebra which contains the cylinders...) The
above can be loosely stated as “any set in M is determined by countably many coordinates”
**Please note the notation used for the box and cylinder sets above.
Answer: To show M ⊂ M∗ , note that πα−1 (Eα ) = β∈A Eβ , where Eβ = Xβ ∀ β 6= α. In this form, it
Q
B
is clear that C ⊂ ⊂ M∗ ⇒ M ⊂ M∗ (by Lemma 1.1).
Next, let us prove that M = M̃:
Proof. Suppose {Fi }∞1 ∈ M̃. Then since M̃ is a union of σ-algebras, it must be that Fi ∈ MA0 for at least
one A0 . Taking A00 to be the union of one of the A0 s which satisfies Fi ∈ MA0 for each i. Thus, A00 will
naturally also be a countable set. Since A00 is a countable set, MA00 ⊂ M̃ ⇒ ∪∞1 Fi ∈ M̃ by Lemma 1.1.
Next, suppose F ∈ M̃, then ∃A0 s.t. F ∈ MA0 , which implies F c ∈ MA0 , and since MA0 ⊂ M̃, ⇒ F c ∈ M̃,
and hence M̃ is indeed a σ-algebra.
Next, since A0 ⊂ A, ⇒ MA0 ⊂ M(= MA ) ∀A0 , and thus since MA0 ⊂ M ∀A0 ⇒ M̃ ⊂ M. To show the
opposite inclusion, we know that ∀ α ∈ A ∃ a countable subset A0 ⊂ A s.t. α ∈ A0 , namely {α}. In this
form, it is perfectly clear that π −1 (Eα ) ⊂ M̃, since π −1 (Eα ) ∈ MA0 ={α} ⇒ M ⊂ M̃. And thus M = M̃.
Let us now turn our attention Q in which the generating family of sets for MA takes. Each set
Q to the form
0
is in the form π −1 (Eα ) = ( β∈A0 Eβ )×( γ∈A\A0 Xγ ), where A0 is a countable set, and Eβ = Xβ ∀β 6= α.
In this form, it is clear that after
Q countably Q many intersections, compliments and unions, ∀E ∈ MA0 ,
E will still be in the form of ( β∈A0 Eβ ) × ( γ∈A\A0 Xγ ), where A0 is a countable set, and Eβ ∈ Mβ .
However, when looking at the boxes, it is clear that ∃E ∈ M(B) s.t. E = ( β∈B Eβ ) × ( γ∈A\B Xγ ),
Q Q
where B is an uncountable set, and Eβ ∈ Mβ .
5
Proposition. 1.4:
Pn
If µ1 , . . . , µn are measures on (X, M) and a1 , . . . , an ∈ [0, ∞), then 1 aj µj is a measure on
(X, M).
n
G Xn n
G ! Xn ∞
X ! Xn ∞
X ! X∞
µ Ei = aj · µj Ei = aj · µj Ei = aj · µj Ei = µ(Ei )
i=1 j=1 i=1 j=1 i=1 j=1 i=1 i=1
If (X, M, µ) is a measure space and {Ej }∞ 1 ⊂ M, then µ(lim inf Ej ) ≤ lim inf µ(Ej ). Also,
µ(lim sup Ej ) ≥ lim sup µ(Ej ) provided that µ(∪∞
1 Ej ) < ∞.
Proof. We first recall the definitions of lim inf and lim sup for a sequence of sets as:
∞ ∞ ∞ ∞
! !
[ \ \ [
lim inf (Fn ) := Fn , and lim sup(Fn ) := Fn
n→∞ n→∞
k=1 n=k k=1 n=k
1 ⊂ M, then:
If {Bi }∞
(a) µ(∩∞
1 Bi ) ≤ µ(B1 ) (or µ(Bk ) by switching B1 for Bk ).
(b) µ(∪∞
1 Bi ) ≥ µ(B1 ) (or µ(Bk ) by switching B1 for Bk ).
Note: I sorta forget these were either covered or corollaries from Thm 1.8 in Folland, hence why I included
it here - oh well (but I did give a slightly more concise proof for (b) :) )
Proof.
(a) Since (∩∞ ∞ ∞ ∞ ∞
1 Bi ) t (B1 \ ∩1 Bi ) = B1 ⇒ µ(B1 ) = µ(∩1 Bi ) + µ(B1 \ ∩1 Bi ) ⇒ µ(∩1 Bi ) ≤ µ(B1 ).
6
We now have all the necessary tools to prove the proposition as follows:
∞ ∞ ∞ ∞
[ \ ∗
\ \ ∗
µ lim inf Ej = µ Ej = lim µ Ej = lim inf µ Ej ≤ lim inf µ (Ej )
k→∞ k→∞ k→∞
k=1 j=k j=k j=k
∗ ∗
Where = is by µ’s “Continuity from below” since ∩∞ ∞
j=k Ej ⊂ ∩j=k+1 Ej ∀k ∈ N, and ≤ is by Lem 2.1 (a).
∞ ∞ ∞ ∞
\ [ ?
[ [ ?
µ lim sup Ej = µ Ej = lim µ Ej = lim inf µ Ej ≥ lim inf µ (Ej )
k→∞ k→∞ k→∞
k=1 j=k j=k j=k
? ?
Where = is by µ’s “Continuity from above” since ∪∞ ∞
j=k+1 Ej ⊂ ∪j=k Ej ∀k ∈ N, and ≤ is by Lem 2.1 (b).
Given a measure space, (X, M, µ) and E ∈ M, define µE (A) = µ(A ∩ E) for A ∈ M. Then µE is
a measure.
1 ⊂ M and
Proof. We first confirm that µE () = 0 since µE () = µ( ∩ E) = µ() = 0. Next, let {Fi }∞
{Fi }∞
1 disjoint. Then:
∞ ∞ ∞ ∞ ∞
! !! !!
∗
[ [ [ X X
µE Fi = µ E ∩ Fi =µ E ∩ Fi = µ (E ∩ Fi ) = µE (Fi )
i=1 i=1 i=1 i=1 i=1
∗
Where = since if {Fi }∞ ∞
1 is a disjoint family of sets, then {Fi ∩ E}1 will be as well. Thus, we have shown
µE is indeed a measure.
7
1.9 Folland 1.13
Prove the following Proposition:
Proposition. 1.8:
Every σ-finite measure is semi-finite.
Proof. Let µ be a σ-finite measure on the measurable space (X, M). Firstly, if µ(X) < ∞, µ will trivially
be semi-finite. Therefore, suppose µ is σ-finite, but not finite. Now, let us arbitrarily pick E ∈ M s.t.
µ(E) = ∞ (we know at least one such element exists, namely X, since otherwise µ would be finite). From
1 ⊂ M s.t. X = ∪1 Fi and µ(Fi ) < ∞ ∀i ∈ N.
the definition of µ being σ-finite, we know that ∃ {Fi }∞ ∞
If µ∗ is an outer measure ∞ ∗
P∞ ∗ on X and {Aj }1 is a sequence of disjoint µ -measurable sets, then
∗ ∞
µ (E ∩ (∪1 Aj )) = 1 µ (E ∩ Aj ) for any E ⊂ X.
Now, let us define Bn := ∪n1 Ej . Now, since Aj is µ∗ -measurable ∀j ∈ N, we know that ∀n > 1:
Therefore, iteratively using the above formula (by induction) for Bn , . . . , B2 , and countable additivity
being trivial for n = 1, we have shown that:
[n n
X
µ∗ E ∩ Aj = µ∗ (E ∩ Aj ) , ∀n ∈ N
j=1 j=1
8
Now, by monotonicty, we can easily see that:
∞
[ n
[ n
X
µ∗ E ∩ Aj ≥ µ∗ E ∩ Aj = µ∗ (E ∩ Aj ) , ∀n ∈ N
j=1 j=1 j=1
S∞ P∞
And hence µ∗ E ∩ j=1 Aj ≥ j=1 µ∗ (E ∩ Aj ). And thus since we shown both ≥ and ≤, we can
P∞ ∗
conclude that µ∗ (E ∩ (∪∞
1 Aj )) = 1 µ (E ∩ Aj ).
Let A ⊂ P(X) be an algebra, Aσ the collection of countable unions of sets in A, and Aσδ the
collection of countable intersections of sets in Aσ . Let µ0 be a premeasure on A and µ∗ the induced
outer measure.
a) For any E ⊂ X and > 0, there exists B ∈ Aσ with E ⊂ B and µ∗ (B) ≤ µ∗ (E) +
b) If µ∗ (E) < ∞, then E is µ∗ -measurable ⇐⇒ there exists C ∈ Aσδ with E ⊂ C and
µ∗ (C\E) = 0.
Proof.
a) Let us recall the definition of µ∗ (E) as:
(∞ ∞
)
X [
1 ⊂ A, E ⊂
∗ {Ai }∞
µ (E) := inf µ0 (Ai ) Ai
i=1 i=1
P∞
Therefore, by the definition of inf, ∀ > 0 ∃{Bi }∞ ∞
1 s.t. E ⊂ ∪1 Bi and
∗
1 µ0 (Bi ) ≤ µ (E) + .
Therefore, if we define B := {Bi }1 (same seq. as before), we note that B ∈ Aσ , and also that:
∞
∞
∗ X
µ∗ (B) ≤ µ0 (Ai ) ≤ µ∗ (E) +
i=1
∗
Where ≤ because µ0 (Bi ) = µ∗ (Bi ), and B is µ∗ -measruable.
b) Let us begin with the forward direction (µ∗ (E) < ∞, and E is µ∗ -measurable). From part (a),
we know ∃ {Ci } ⊂ Aσ s.t. E ⊂ Ck and µ∗ (Ck ) ≤ µ∗ (E) + k1 ∀k ∈ N. Let us now define
C := ∩∞ 1 Ci , to which we notice that C ∈ Aσδ and E ⊂ C since E ⊂ Ck ∀k ∈ N, and hence
µ∗ (E) ≤ µ∗ (C). Furthermore, we note that since Ck is µ∗ -measurable, so too will Ckc , and hence
∪∞ c ∞ c c ∗ ∗
1 Ci = (∩1 Ci ) = C is µ -measurable, and hence C is µ -measurable. Now, the following
observation becomes apparent:
∞
! n
!
\ \
∗ ∗ ∗
µ (C) = µ Ci = lim µ Ci ≤ lim µ∗ (Cn ) = µ∗ (E)
n→∞ n→∞
i=1 i=1
9
Moreover, using the fact that E ⊂ C the above now actually implies that µ∗ (E) = µ∗ (C). We also
recall that since E c is µ∗ -measurable, and since we already showed that C was µ∗ -measurable, we
can now also say that C ∩ E c = B\E is µ∗ -measurable, and also note that hence:
For the backward direction (there exists C ∈ Aσδ with E ⊂ C and µ∗ (C\E) = 0), first note that
since E ⊂ C), C = (B\E)∪E. Next, since µ∗ is the Carathèodory extension, C\E is µ∗ -measurable.
Therefore, we can easily conclude that E = B\(B\E) is also µ∗ -measurable.
P∞ P∞
Where µ∗ (C\E) = µ∗ (∪∞ 1 Ci \Ei ) ≤ 1 µ∗ (Ci \Ei ) = 1 0 = 0. And hence µ ∗ (E) < ∞ did not
matter if µ0 is σ-finite.
If E ∈ Mµ and µ(E) < ∞, then ∀ > 0 ∃ a set A that is a finite union of open intervals such that
µ(E4A) < ∞.
Proof. We recall that by Theorem 1.18, ∃Uopen s.t. E ⊂ U and µ(U) ≤ µ(E) + 12 . Furthermore, by the
inequality just stated, we know that µ(U), µ(E) < ∞, and hence:
1
µ(U\E) = µ(U) − µ(E) <
2
∞
X
µ (Ui ) = µ(U) < µ(E) < ∞
i=1
10
P∞
P∞ the series 11 µ (Ui ) must converge, and hence, by the definition of convergent series’, ∃N ∈ N
Therefore,
s.t. N +1 µ (µ(Ui )) < 2 , and thus the inequality we sought to prove has now been shown.
Carrying on, let us define Ũ := {Ui }N
1 . Since Ũ ⊂ U ⇒ µ(Ũ) ≤ µ(U) < ∞ and also ⇒ Ũ\E ⊂ U\E,
hence:
1
µ(Ũ\E) ≤ µ(U\E) <
2
Now, also since µ(Ũ) < ∞, and since Ũ ⊂ U ⇒ E\Ũ ⊂ U \Ũ , we can see that:
∞
X 1
µ(E\Ũ) ≤ µ(U\Ũ) = µ(U) − µ(Ũ) = µ(Ui ) <
2
i=N +1
Therefore, by combining the last two main inequalities, we have found a set A = Ũ which is a finite union
of open intervals such that:
1 1
µ(E4Ũ) = µ(E\Ũ) + µ(Ũ\E) < + =
2 2
Proof.
a) We first note that we may construct {a} from a countable intersection of h-intervals as follows:
∞
\
{a} = (a − 1/n, a]
n=1
Furthermore, since (a − 1/n, a] ⊃ (a − 1/(n + 1), a] ∀n ∈ N, we may invoke continuity from above
in that:
∗
µF ({a}) = lim µF ((a − 1/n, a]) = lim (F (a) − F (a − 1/n)) = F (a) − F (a−)
n→∞ n→∞
∗
Where = can be rigorously shown by noting that since F is an increasing function:
11
b) We first note that we may construct [a, b) from a union of countable intersections and unions of
h-intervals as follows:
∞ ∞
! !
\ [
[a, b) = [a, (a + b)/2] ∪ (a, b) = (a − 1/n, (a + b)/2] ∪ (a, b − 1/m]
n=1 m=1
Therefore, since all the sets we’ve been dealing with so far have been bounded, we can see now that:
µF [a, b) = µF [a, (a + b)/2] + µF (a, b) − µF (a, b) ∩ [a, (a + b)/2]
= µF [a, (a + b)/2] + µF (a, b) − µF (a, (a + b)/2]
h i h i h i
= F (a + b)/2 − F a − + F b − − F a − F (a + b)/2 − F a
h i h i h i
= F (b−) − F (a−) + F (a + b)/2 − F (a + b)/2 + F a − F a
= F (b−) − F (a−)
c) We first note that we may construct [a, b] from countable intersection of h-intervals as follows:
∞
\
[a, b] = (a − 1/n, b]
n=1
Thus, by making the change of variables of (a + b)/2 → b, from the first half of b), we have already
shown that µF ([a, b]) = F (b) − F (a−).
d) We first note that we may construct (a, b) from a countable union of h-intervals as follows:
∞
[
(a, b − 1/n]
n=1
Thus, from the second half of b), we have already shown that µF ((a, b)) = F (b−) − F (a).
12
1.14 Folland 1.30
Prove the following Proposition:
Proposition. 1.13:
If E ∈ L and m(E) > 0, for any α < 1 ∃ an open interval Iˆ such that m(E ∩ I) > αm(I).
Proof. If α ≤ 0, since m(E) > 0 ⇒ ∃F ⊂ E s.t. m(F ) > 0, and F = (a, b], a < b. If we thus take:
ˆ 1 3
I= (a + b), (a + b)
4 4
Now, for the sake of contradiction, assume ∀I = (a, b), a < b, we have: m(E ∩ I) ≤ αm(I). Let us choose
1 > 0 so that 1 < 1−α α (and hence α(1 + 1 ) < 1). Moreover,
P∞ from (Folland) Theorem 1.18, we know
that ∀2 > 0 ∃I = t∞
1 (ai , bi ) s.t. E ⊂ I and m(I) = 1 m (ai , bi ) < m(E) + 2 . Next, from our
discussion on (a, b) v.s. (a, b], we can actually write I = t∞
1 (ai , bi ], where I still satisfies everything that
it did beforehand. Now, if we let 2 = m(E)1 , (which we can certainly do since m(E) < ∞), we see that:
∞
X 1−α 1
m(I) = (ai , bi ] < m(E) + m(E)1 = m(E)(1 + 1 ) < m(E) 1 + = m(E)
i=1
α α
⇒ αm(I) < m(E)
Therefore, by combining the above inequality with our assumption in that m(E ∩ Ik ) ≤ αm(Ik ) ∀k ∈ N,
and that E ⊂ I, we see that:
∞
X ∞
X
m(E) = m(E ∩ I) = m(E ∩ Ii ) ≤ αm(Ii ) = αm(I) < m(E)
i=1 i=1
Which is obviously a contradiction on the requirement of m(E) > 0, hence the converse must be true:
I.e. our Proposition is true.
Proof. From (Folland) 1.30, we know that ∃ I s.t. 34 m(I) < m(E ∩ I). Let us now define F := E ∩ I ⊂ E,
and naturally we will have {F − F } ⊂ {E − E}, hence if ∃ an interval centered at 0 in {F − F }, so too
will that interval be in {E − E}.
13
We now claim that F ∩ {F + x0 } 6= ⇒ x0 ∈ {F − F }. To see this, let y ∈ F ∩ {F + x0 } ⇒ y ∈
F and ∃x ∈ F s.t. y = x + x0 ⇒ x0 = y − x, y, x ∈ F ⇒ x0 ∈ {F − F }.
Trivially 0 ∈ {F − F } since F 6= . Let us now let z0 ∈ R s.t. |z0 | < 12 m(I) < 34 m(I) < m(F ). If we can
show that F ∩ {F + z0 } 6= ⇒ − 12 m(I), 12 m(I) ⊂ {E − E}. Therefore, the remainder of this proof
will be dedicated to showing F ∩ {F + z0 } =6 where |z0 | < 12 m(I).
Firstly, we note that:
3 1
m(I\F ) = m(I) − m(F ) = m(I) − m(E ∩ I) ≤ m(I) − m(F ) = m(F )
4 4
Furthermore, by applying the useful fact that A ∩ B = ((A\C) ∩ B) ∪ ((C\A) ∩ B) twice, we find:
I ∩ {I + z0 } = F ∩ {F + z0 } ∪ F ∩ ({I\F } + z0 ) ∪ (I\F ) ∩ {I + z0 }
Our strategy now will be to show that m F ∩ {F + z0 } > 0, which therefore would imply I ∩ {I + z0 }
also has positive measure, and hence cannot be empty. To see this first note the following four properties:
m(I ∩ {I + z0 }) ≤ m F ∩ {F + z0 } + m F ∩ ({I\F } + z0 ) + m (I\F ) ∩ {I + z0 }
1
and: m F ∩ ({I\F } + z0 ) ≤ m {(I\F ) + z0 } = m I\F ] ≤ m(F ) from previously
4
1
and: m F ∩ ({I\F } + z0 ) ≤ m(I\F ) ≤ m(I) again
4
1
and: m(I) < m(I) − |z0 | = m I ∩ {I + z0 }
2
And hence combing all these we see that:
1 1
m(I) < m I ∩ {I + z)} ≤ m F ∩ {F + z0 } + m(I) ⇒ m F ∩ {F + z0 } > 0
2 2
There exists a Borel set A ⊂ [0, 1] such that 0 < m(A ∩ I) < m(I) for every sub-interval I of [0, 1].
(Hint: Every sub-interval of [0, 1] contains Cantor-type sets of positive measure.)
Proof. The first observation we need to make is that since |Q| = ℵ0 ⇒ |Q × Q| = ℵ0 (ℵ0 := “countably
infinite”). Therefore, we can actually write the set of all closed sets Ik inside [0, 1] where Ik ’s endpoints
are rational numbers as a countable list: Iˆ = {Ij }∞1 . By the hint, we know that every sub-interval of
[0, 1] contains Cantor-type sets (which will certainly have rational endpoints). Our plan will therefore
be through induction, to explicitly describe a Borel set made up of necessary Cantor-like sets which will
satisfy the needed inequality.
Let Ak , Bk be strict subsets of Ik (which we can do because we’re assuming I 6= , and due to the density
of the rationals) s.t. Ai ∩ Bk = and m(Ai ), m(Bj ) > 0 ∀i, j ≤ N . We can therefore define:
N
G
CN := IN \ (Aj ∪ Bj )
j=1
14
And therefore, we can find a Cantor-type set DN and D̃N s.t. m(DN ), D(D̃N ) > 0 ∀N ∈ N. If we let
D := ∪∞
N =1 DN , then ∀ sub-intervals I ⊂ [0, 1], ∃N s.t. IN ⊂ I and we will have:
2 Chapter 2
2.1 Folland 2.1
Prove the following Proposition:
Proposition. 2.1:
Proof. To be clear on notation, if X = {±∞}, then either X = {∞} or X = {−∞}, and naturally
{−∞, ∞} = 6 X.
For the forward direction, since f is measurable and {±∞} ∈ BR , it implies f −1 ({±∞}) ∈ M. Fur-
thermore, again by f ’s is measurability and since R ∈ BR , it implies f −1 (R) ∈ M. Therefore, we may
conclude that if B ∈ BR , then f −1 (B) ∈ M and f −1 (B)∩f −1 (R) = f −1 (B)∩Y ∈ M, I.e., f is measurable
on Y .
For the converse, if we let B ∈ BR , then we can see that:
And since f −1 (R) is measurable, naturally f −1 (B) ∩ f −1 (R) = f −1 (B ∩ R) is as well. Next, we note that:
b) Fix a ∈ R, and define h(x) = a if f (x) = −g(x) = ±∞, and h(x) = f (x) + g(x) otherwise.
Then h is measurable.
15
Proof. We actually do this problem in reverse ordering.
b) We prove this fact by separating the problem into 2 lemmas, and one final main result:
For the first mini-lemma, we note that A∞ := {x ∈ X | f (x) = −g(x) = ±∞} is measurable since
f and g are measurable.
For the second mini lemma, we make the observation that:
h−1 ({∞}) = (f + g)−1 ({∞}) = f −1 ({∞}) ∩ g −1 ((−∞, ∞]) ∪ f −1 ((−∞, ∞]) ∩ g −1 ({∞})
Since h(x) = ∞ ⇐⇒ either [f (x) = ∞ and g(x) > −∞] or [g(x) = ∞ and f (x) > −∞], or [
f (x) = g(x) = ∞]. Similarly for the {−∞} (sub-) case:
h−1 ({−∞}) = (f +g)−1 ({−∞}) = f −1 ({−∞})∩g −1 ([−∞, ∞)) ∪ f −1 ([−∞, ∞))∩g −1 ({−∞})
Since h(x) = −∞ ⇐⇒ either [f (x) = −∞ and g(x) < ∞] or [g(x) = −∞ and f (x) < ∞], or [
f (x) = g(x) = −∞]. We naturally recognize the above to certainly be measurable (again) since f
and g are measurable.
Now for the final main result. Let b ∈ R, then:
h−1 ((b, ∞]) = h−1 ((b, ∞)) ∪ h−1 ({∞)}
Since we already showed that h−1 ({∞}) is measurable, we now seek to show that h−1 ((b, ∞)) is
measurable. This can be seen since:
(
−1 (f + g)−1 ((b, ∞)) if a ≤ b
h ((b, ∞)) =
(f + g)−1 ((b, a)) ∪ (h)−1 ({a}) ∪ (f + g)−1 ((a, ∞)) if a > b
(
(f + g)−1 ((b, ∞)) if a ≤ b
= −1
A∞ ∪ (f + g) ((b, ∞)) if a > b
Where we already showed that A∞ is measurable, and by f and g’s measurability, all the sets
above which make up h−1 ((b∞)) are measurable, and hence h−1 ((b, ∞]) is measurable; therefore,
h is measurable.
a) Let us define Q+ := {r ∈ Q | r > 0} and Q− := {r ∈ Q | r < 0}, which is a subsets of Q and hence
countable. Suppose now that f, g ≥ 0, if a ≥ 0, then we will have:
(f g)−1 ((a, ∞]) = {x ∈ X | f (x)g(x) > a}
[
= {x ∈ X | f (x) > r} ∩ {x ∈ X | g(x) > a/r}
r∈Q+
Now, by our previous work, since f + , g + , f∗− , g∗− ≥ 0, it follows that the first half of the above ex-
pression is measurable (since by part b, we showed that the addition of two measurable functions as
defined in this question is measurable). And also recalling that f measurable ⇐⇒ −f measurable,
we can therefore conclude that f g is indeed measurable.
16
2.3 Folland 2.3
Prove the following Proposition:
Proposition. 2.3:
If {fn } is a sequnce of measurable functions on X, then {x | lim fn (x) exists} is a measurable set.
Proof. We first recall that by (Folland) Proposition 2.7, when {fn } is defined as in the question, g3 (x) =
lim supn→∞ fn (x) and g4 (x) = lim inf n→∞ fn (x) are both measurable. If, as in Exercise 2.2, we let a = 1,
then function g3 − g4 is measurable (and is equal to 1 when g3 = g4 = ±∞). Finally, by noting that
lim fn (x) exists ⇐⇒ g3 = g4 , we can actually write:
Which is most certainly measurable since g3 and g4 are measurable, and the difference of such measurable
functions is also measurable (Corollary of Exercise 4.2 by combining the fact that f measurable ⇐⇒
−f measurable, and taking f − g = f + (−g)).
and since: [
f −1 ((a, ∞]) ⊂ f −1 ((r, ∞]) ∈ M
r∈Q+
a
Proof. We claim that f (x) := inf{α ∈ R | x ∈ Eα }, where Eα has the same construction as given in the
Proposition, will satisfy the requirements of being measurable and the stated inequalities. We begin first
by showing the latter.
Suppose x ∈ Eα , then by the construction of f , we immediately have f (x) ≤ α. Now, suppose α ∈ Eαc ,
then ∀β ≤ α, Eαc ⊂ Eβc since Eβ ⊂ Eα ; therefore, x ∈ Eβc ⇒ x 6∈ Eβ ∀β ≤ α ⇒ f (x) ≥ α if x ∈ Eαc .
17
Again by the construction of f , it is clear that ∪α∈R Eα = X and ∪α∈R Eαc = X. From this, given ∀x ∈ X,
we know that ∃ α, β ∈ R such that x ∈ Eα and x ∈ Eβ and most importantly since α, β ∈ R:
And hence f (x) 6= ±∞ irregardless of x. It’ll now be a lot easier to conclude measurability since we no
longer have to worry about the possibility that f (x) = ±∞.
Let us now take r ∈ Q, and note that by first set of inequalities established, if x ∈ X, then f (x) < r ⇐⇒
∃q ∈ R s.t. x ∈ Eq . Equivalently: f −1 ((−∞, r)) = ∪q<r Eα . By the density of Q, we can actually restrict
that q, r ∈ Q. We therefore have:
[
f −1 ((−∞, r)) = Eq , where q, r ∈ Q
q<r
And since Eq ∈ M ∀q, and since {q ∈ Q | q < r} is a countable set, we naturally have f −1 ((−∞, r)) ∈ M.
Furthermore, by the inequalities established, we also have:
[
f −1 ([r, ∞)) = Erc ∈ M
q<r
Proof. We first state our strategy: If we can show that ∀a ∈ R, f −1 ([am∞)) is an interval, then f must
be Borel measurable., let us note that as trivial corollary of (Folland) Proposition 2.3, f measurable
⇐⇒ −f measurable. Thus, without loss of generality, assume f is monotone increasing. Suppose now
that a ∈ R, x ∈ f −1 ([a, ∞), and y ∈ [x, ∞). Therefore, since f is monotone increasing:
Since this is true ∀x, y ∈ [a, ∞), it actually proves that f −1 ([a, ∞)) is indeed an interval, and therefore
Borel measurable, and hence f is Borel measurable since this is true ∀a ∈ R.
18
Proposition. 2.7:
Let f : [0, 1] → [0, 1] be the Cantor Function (Folland Section 1.5), and let g(x) = f (x) + x.
a) g is a bijection from [0, 1] to [0, 2], and h = g −1 is continuous from [0, 2] to [0, 1].
b) If C is the Cantor set, m(g(C)) = 1.
Proof.
a) We first recall (from Folland) that the Cantor Function, f (x) is monotone increasing, and naturally
h(x) = x is a strictly increasing function, and hence g(x) = f (x) + x is also strictly increasing
and therefore injective. Next, to show surjectivity, note that g is a continuous function, and
g(0) = f (0) + 0 = 0, and g(1) = f (1) + 1 = 2; hence, by the intermediate value theorem, g is
surjective.
We now have all the necessary components to conclude that g is a bijection, and since g is a
continuous bijective function, and [0, 1] is compact, g’s inverse, g −1 , is continuous from [0, 2] to
[0, 1].
b) Firstly, by g’s surjectivity, and C being measurable, we see that:
Next, since C is a closed set ⇒ [0, 1]\C is an open set. Therefore, since all open subsets of [0, 1] may
be written as a countable union of disjoint open sets, let us write [0, 1]\C = t∞ 1 Oj , Oj = (aj , bj ).
Now, since f is by construction constant on [0, 1]\C, and recalling that m(C) = 0 ⇒ m([0, 1]\C) =
1 ⇒ m(t∞ 1 Oj ) = 1, we see:
∞
G X∞
m(g([0, 1]\C)) = m g Oj = m(g(Oj ))
j=1 j=1
∞
X
= m(f (bj ) − f (aj )) + m(bj − aj )
j=1
X∞
= m(Oj ) since f (bj ) = f (aj ) ∀j ∈ N
j=1
∞
G
=m Oj
j=1
=1
c) To show Lebesgue measurability, naturally B ⊂ C, and since C is measurable with measure m(C) =
0, it implies m(B) ≤ m(C) = 0, and hence Lebesgue measurable since null sets are measurable.
19
For the sake of contradiction, suppose B = g −1 (A) is Borel measurable. In part a), we showed that
g −1 is continuous and bijective; therefore g(B) = g(g −1 (A)) = A. However, by the continuity of g,
if g −1 (A) was Borel, so too would g(g −1 (A)) = A, hence a contradiction since A is not Lebesgue
measurable; therefore, B cannot be Borel measurable.
(
1 if x ∈ B
d) Let F = χB ; I.e., F (x) = , and also set G = g −1 . Naturally G is Lebesgue mea-
0 if x ∈ B c
surable since it is continuous, we now wish to prove that so too is F . This can be seen by notic-
ing F −1 ((a, ∞)) = or B or R, but all these possibilities are Lebesgue measurable, hence F is
Lebesgue measurable. We can now look at the following reasoning:
(F ◦ G)−1 ((1/2, ∞)) = G−1 ◦ F −1 ([1/2, ∞)) = {x ∈ [0, 2] | χB (g −1 (x)) ∈ [1/2, ∞)}
= {x ∈ [0, 2] | g −1 (x) ∈ B}
= G−1 (B) = g(g −1 (A)) = A
Now since A is not Lebesgue measurable, F ◦ G also will not be Lebesgue measurable.
Proof.
a) For the forward direction, suppose a) holds. Then let N ∈ M be a measurable set s.t. µ(N ) = 0,
and N1 ⊂ N . If we define f :≡ 0 and χN1 := 1 if x ∈ N1 , and 0 otherwise, then trivially f is
measurable and f = χN1 µ-a.e., so by our assumptions g is measurable. Now, by noting that
χ−1
N1 ({1}) = N1 ∈ M by g’s measurability, and since this is true ∀N1 ⊂ N , we have arrived at the
definition of µ being complete.
For the backward direction, suppose µ is complete, and let f be measurable and f = g µ-a.e.
Explicitly, let N ∈ M be the measurable set s.t. µ(N ) = 0 and f (x) = g(x) ∀x ∈ N c . Then if A is
measurable, we have:
Looking at the right hand side, we can see g −1 (A) ∩ N ⊂ N is measurable by the definition of µ
being a complete measure since µ(N ) = 0. Furthermore, f −1 (A)\N ⊂ f −1 (A) since f is measurable.
With these two facts, we may therefore conclude that g is indeed measurable.
b) For the forward direction, suppose b) holds. Then let N ∈ M be a measurable set s.t. µ(N ) = 0,
and N1 ⊂ N . If we let fn = 0 and χN1 as before, then like in the forward direction of a), we have
fn → χN1 µ-a.e., so χN1 is measurable. Therefore, χ−1N1 ({1}) ∈ M, and since this is true ∀N1 ⊂ N ,
we have arrived at the definition of µ being complete.
20
For the backward direction, suppose µ is complete, and fn is measurable ∀n ∈ N, and fn → f
µ-a.e. By (Folland) Proposition 2.7, g3 (x) = lim supj→∞ fj (x) is measurable since fn is measurable
∀n ∈ N. Furthermore, since fn → f µ-a.e., we have g3 = f µ-a.e., and thus by the backward
direction of part a) above, f is measurable.
If f ∈ L+ and f < ∞, then {x | f (x) = ∞} is a null set and {x | f (x) > 0} is σ-finite.
R
Suppose {fn }∞ +
R R R R
1 ⊂ L , fn → f pointwise,R and f = lim fn < ∞. Then E f = lim E fn
∀E ∈ M. However, this need not be true if f = lim fn = ∞.
21
Furthermore, by (Folland) Theorem 2.15, we have:
Z Z Z Z
f = (χE f + χE c f ) = χE f + χE c f
∗ R R R ∗∗ R R R
Where we R have = sinceR E f = χE f + χE c f , and = since lim inf fn = lim fn = f and
lim inf − g = − lim sup g. However, since all terms above are finite, we may gain by apply Fatou’s
Lemma (and in noticing the similarity to the steps made above) to see that:
Z Z Z Z Z Z Z
lim sup fn = lim sup fn − fn = f − lim inf fn ≤ f − f
n→∞ Ec n→∞ E n→∞ E E
And therefore all the inequalities in the equation(s) above are actually equalities, and so we have:
Z Z Z Z
lim inf fn = lim sup fn = lim f= f
n→∞ E n→∞ E n→∞ E E
R R
We now turn our attention showing the above result need not hold if f = lim f = ∞ by means of a
counter-example. Let E = (0, 1], f = χ[2,∞) , and fn = χ[2,∞) + nχ(0,1/n] . Then fn → f p.w., and:
Z Z
fn = nµ((0, 1/n]) = 1 ∀n ∈ N ⇒ lim fn = 1
(0,1] n→∞ (0,1]
R R R R R
However, (0,1]
f = 0, thus E
f = lim E
fn need not be true if lim f= f = ∞.
Proof. Trivially, since f ∈ L+ , we have that λ(E) ≥ 0 ∀E ∈ M. Moreover, one can see that λ() = 0:
Z Z
λ() = f dµ = χ f dµ = 0
22
To fully show
P∞that λ is a measure on M, we need that for any disjoint sequence of sets, {Ej }∞
1 ∈ M,
∞
λ(t1 Ej ) = 1 λ(Ej ). We can deduce this fact from the following:
∞
G Z Z
λ Ej = f dµ = χ(t∞
1 Ej )
f dµ
j=1 t∞
1 Ej
∞
Z X ∞ Z
∗ ∗
X
= χEj f dµ = χEj f dµ = by (Folland) Theorem 2.15
j=1 j=1
∞ Z
X ∞
X
= f dµ = λ(Ej )
j=1 Ej j=1
We have thus shown all the necessary conditions for λ to be a measure do indeed hold.
Pn
Next, let g ∈ L+ , and assume that g is simple ⇒ g = 1 aj χEj . Therefore:
Z n
X n Z
X n Z
X
g dλ = aj λ(Ej ) = f dµ = χEj f dµ
j=1 j=1 Ej j=1
n
Z X Z
∗ ∗
= aj χEj f dµ = gf dµ = by (Folland) Theorem 2.15
j=1
And so we get the required result when g is simple. However, by (Folland) Theorem 2.10, we know that
since f ∈ L+ , ∃{φn }∞
1 s.t. 0 ≤ φ1 ≤ φ2 ≤ · · · ≤ f , φn → f p.w., and φn → f uniformly on any set on
?
which f is bounded. Therefore, we can apply the Monotone Convergence Theorem (used if = denoted)
as follows:
Z Z Z Z
? ∗ ? ∗
g dλ = lim φn dλ = lim φn f dµ = gf dµ = since φn simple ∀n ∈ N
n→∞ n→∞
Proof. Firstly, By (Folland) Exercise 2.12 (proved above - 5.2), we know that F := {x |f (x) > 0} is
σ-finite. In the proof of (Folland) 2.12, we showed that Fn := {x | f (x) > 1/n} has the nice properties
of µ(Fn ) < ∞ and ∪∞ 1 Fn = F . Furthermore, it is also apparent from the construction of Fn that
Fn ⊂ Fn+1 ∀n ∈ N - I.e., {Fn }∞ ∞
1 is monotone increasing, and so {χFn }1 will be an increasing sequence
+
in L s.t. χFn ≤ χFn+1 ∀n ∈ N, and limn→∞ χFn = χF .
Since {χRFn }∞
1 Rand χF satisfy necessary conditions for the Monotone Convergence Theorem, and in
noticing f = χF f , we may apply it as follows:
Z Z Z Z
f = χF f = lim χFn f = f
n→∞ Fn
23
We also note that, since χFn ⊂ χF ∀n ∈ N, we have:
Z Z Z Z
f = χF f ≤ χFn f = f ∀n ∈ N
Fn
R R
Therefore, Fn is an increasing sequence with the limit of f . So by this convergence, we have ∀ > 0,
∃N ∈ N such that: Z Z
f> f −
FN
{fn }∞
R
However,
R since 1is monotone increasing with the limit of f , we have fn ≤ f ∀n ∈ N ⇒ fn ≤
f ∀n ∈ N. And hence taking the lim sup on both sides, we get:
Z Z Z
lim sup fn ≤ lim sup f = f
n→∞ n→∞
Which can be true ⇐⇒ all the inequalities above are actually equalities, hence we have:
Z Z Z Z
lim fn = lim sup fn = lim inf fn = f
n→∞ n→∞ n→∞
Proof. Firstly, we note that by (Folland) Corollary 2.2, since f ∈ C 1 (R) ⇒ f ∈ C(R), we have that f is
Borel measurable.
24
Next, we prove that gn := f (x + 1/n) is Borel measurable. This is actually quite easy since hn = x + 1/n
is naturally Borel measurable, and hence f ◦ hn = gn is Borel measurable since both f and gn are Borel
measurable.
Next, since f ∈ C 1 (R), we know that limh→0 f (x+h)−f
h
(x)
= f 0 (x), f 0 (x) ∈ R. Therefore, we can also say
0
that limn→∞ n(f (x + 1/n) − f (x)) = f (x) ∀x ∈ R. Since we already showed f (x + 1/n) and f (x) are
Borel Measurable, by (Folland) Proposition 2.6, fn0 := n(f (x + 1/n) − f (x)) is Borel measurable ∀n ∈ N.
Finally, by (Folland) Proposition 2.7, we can conclude that f 0 (x) = limn→∞ fn0 (x) is Borel measurable
since f ∈ C 1 (R), fn0 → f , and {fn }∞
1 is a sequence of Borel measurable functions.
(A generalizedR Dominated
R Convergence
R Theorem)
R If fn , gn , f, g ∈ L1 , fn → f and gn → g a.e.,
fn ≤ gn and gn → g, then fn → f . (Rework the proof of the dominated convergence
theorem).
Proof. By the same reasoning as in Folland, WLOG we may assume fn and f are real-valued, and that
gn + fn ≥ 0 a.e., and gn − fn ≥ 0 a.e. Now, we apply (Folland) Corollary (of Fatou’s Lemma) 2.19 to
both gn + fn and gn − fn as follows (we can do so due to the convergent and L1 assumptions):
Z Z Z Z Z
(g + f ) = lim(gn + fn ) ≤ lim inf (gn + fn ) = g + lim inf fn
Z Z Z Z Z
(g − f ) = lim(gn − fn ) ≤ lim inf (gn − fn ) = g − lim sup fn
And so: Z Z Z Z Z Z Z Z
lim sup fn − g≤− g+ f and g+ f≤ g + lim inf fn
And since Rf, fn ∈ RL1 , we know that the above inequalities imply equalities, everywhere, I.e., lim
R
fn
exists and fn → f .
25
R
Proof. For the forward direction, assume |fn − f | → 0, then since:
Z Z Z Z Z
|fn | − |f | = |fn | − |f | ≤
|f
n | − |f | ≤ |fn − f |, since |fn | − |f | ≤ |fn − f |
we
R know that
R the right hand
R sideR→ 0 as n → ∞, and since the above holds ∀n ∈ N (and since fn , f ∈ L1 ),
|fn | − |f | → 0 ⇒ |fn | → |f |.
R R
For the backward direction, assume R|fn | →R |f |. If we let gnR := |fn | + |f |, then naturally |fn − f | ≤ gn ,
and since fn , f ∈ L1 , we know that gn = (|fn | + |f |) = 2 |f |. We may now invoke the generalized
dominated convergence theorem, (Folland) Exercise 2.20 above, which implies:
Z Z
lim |fn − f | = lim |fn − f |
R
And since fn → f , we therefore have |fn − f | → 0.
Before we begin, we present Folland Exercise 33 as a necessary Lemma for part a):
2.18.1 Folland 2.33
Lemma. 2.1:
R R
If fn ≥ 0 and fn → f in measure, then f ≤ lim inf fn .
Proof.
a) By (Folland) Theorem 2.30, ∃ a subsequence {fnk }∞ 1 s.t. fnk → h, where f = h a.e. Furthermore,
by (Folland) Proposition 2.11, since f = h a.e., f is measurable. As is standard by this point, we
may assume fn and g are real-valued functions; therefore, g + fn ≥ 0 a.e., and g − fn ≥ 0 a.e.
Moreover, we naturally have g + fn → g + f , g − fn → g − f in measure. We now make use of our
Lemma as follows:
Z Z Z Z Z Z
g+ f= (g + f ) ≤ lim inf (g + fn ) = g + lim inf fn
Z Z Z Z Z Z
g− f= (g − f ) ≤ lim inf (g − fn ) = g − lim sup fn
26
Which we recognizeR as in previous
R exercises to be true ⇐⇒ all the above inequalities are actually
equalities; hence: f = lim fn .
b) From (Folland) Proposition 2.29, we know that since fn → f in L1 , fn → f in measure as well.
Thus, since:
µ x ∈ X ||fn (x) − f (x)| − 0| ≥ = µ x ∈ X |fn (x) − f (x)| ≥ → 0 as n → ∞
Hence fn → f in L1 .
Proof. For the forward direction, suppose fn → f in measure (µ a counting measure on N). Then ∀ > 0
∃N ∈ N s.t. ∀n ≥ N :
1
µ x ∈ N |fn (x) − f (x)| ≥ < ⇒ x ∈ N |fn (x) − f (x)| ≥ =
2
I.e., |fn (x) − f (x)| < ∀x ∈ N (and n ≥ N ), which is by definition uniform convergence.
27
For the converse, suppose fn → f uniformly (again µ a counting measure on N). Then ∀ > 0, ∃N ∈ N
s.t. ∀n ≥ N :
|fn (x) − f (x)| < ∀x ∈ N ⇒ µ x ∈ N |fn (x) − f (x)| ≥ = µ() = 0
For which the latter equality vacuously satisfies our definition of convergence in measure.
R R
Proof. We begin by making the following two observations: ∀x ∈ [0, 1], χD dν(y) = {x} dν(y) =
R R
ν({x}) = 1 and ∀y ∈ [0, 1], χD dµ(x) = {y} dν(x) = µ({y}) = 0. Therefore, we’ll now be able to
RR RR
compute χD dµdν and χD dνdµ as follows:
Z Z Z Z Z
χD dµdν = χD (x, y)dµ(x) dν(y) = 0dν(y) = 0
Z Z Z Z Z
χD dνdµ = χD (x, y)dν(y) dµ(x) = dµ(x) = 1
[0,1]
28
2.23 Folland 2.48
Prove the following Proposition:
Proposition. 2.20:
R RR RR
Then |f |d(µ × ν) = ∞, and f dµdν and f dνdµ exist and are unequal.
Proof. We first claim that µ × µ is also a counting measure. We may actually note that fundamentally,
a counting measure τ on N × N will satisfy τ (A × B) = |A||B| = µ(A)ν(B). Therefore, since rectangles
generate the product σ-algebra, the σ-finitness of τ implies µ × ν = τ .
We now proceed to computing each of the quantities of interest. For the first, if we let E := ∪∞
1 {{(n, n)}∪
{(n, n + 1)}}, then clearly |E| = ∞, and |f | = χE . Therefore:
Z
|f |d(µ × ν) = |E| = ∞
Prove (Folland) Theorem 2.38 by using Theorem 2.37 and Proposition 2.12 together with the
following lemmas:
a) If E ∈ M × N and µ × ν(E) = 0, then ν(Ex ) = µ(E y ) = 0 for a.e. x and y.
R f is L-measurable
b) If R y and f = 0 λ-a.e., then fx and f y are integrable for a.e. x and y, and
fx dν = f dµ = 0 for a.e. x and y. (Here the completeness of µ and ν is needed.)
Proof.
a) Immediately by (Folland) Theorem 2.36, since (X, M, µ) and (Y, N, ν) are σ-finite measure spaces,
we have: Z Z
0 = µ × ν(E) = ν(Ex )dµ(x) = µ(E y )dν(y)
29
b) Let us define F := {(x, y) ∈ M × N | f (x, y) 6= 0}. Thus, ∃E ∈ M ⊗ N where µ × ν(E) = 0 and
F ⊂ E. From a), µ(Ex ) = ν(E y ) = 0 for x, y a.e. Furthermore, by the fact that Fx ⊂ Ex and
F y ⊂ E y , by linearity of measures we have µ(Fx ) = ν(F y ) = 0 as well. We may now conclude thus
that: Z Z Z Z
|fx |dν = χFx |fx |dν = 0 = χF y |f y |dµ = |f y |dµ = 0 for a.e. x and y
3 Chapter 3
3.1 Folland 3.2
Prove the following Proposition:
Proposition. 3.1:
Proof.
a) For the forward direction, suppose ν is a signed measure and that E is ν-null. Suppose for the sake
of contradiction that |ν|(E) = ν + (E) + ν − (E) > 0, where ν = ν + − ν − is the Jordon Decomposition
of ν. By the Hahn Decomposition Theorem, ∃ P, N s.t. ν(X) = ν(P ∪ N ) = ν + (P ) − ν − (N ) and
ν + (N ) = 0 = ν − (P ).
We thus can thus make the following observations:
For the converse, suppose |ν|(E) = 0, hence |ν|(E 0 ) = 0 ∀E 0 ⊂ E (E 0 measurable). Since |ν|(E 0 ) =
ν + (E 0 ) + ν − (E 0 ) = 0 ⇐⇒ ν + (E 0 ) = 0 = ν − (E 0 ), we thus trivially satisfy ν(E 0 ) = ν + (E 0 ) −
ν − (E 0 ) = 0 since both are already zero.
b) Let us recall that, explicitly, if ∃ E, F ∈ M s.t. E ∩F = , E tF = X and µ(E 0 ) = 0 = ν(F 0 ) ∀E 0 ⊂
E, F 0 ⊂ F (E 0 , F 0 measurable), we denote this property as ν ⊥ µ.
We begin by showing ν ⊥ µ ⇒ |ν| ⊥ µ. To see this, since F is ν-null, by Part a), we know that
|ν|(F ) = 0. Since |ν| is a positive (regular) measure, by monotonicity we have that |ν| is F -null.
Thus, by definition, |ν| ⊥ µ.
30
We now show |ν| ⊥ µ ⇒ ν + ⊥ µ and ν − ⊥ µ. Since ν = ν + + ν − , we have ν + ≤ ν and ν − ≤ ν,
and so ν + (F 0 ) = ν − (F 0 ) = 0, I.e., F is both ν + -null and ν − -null; hence, ν + ⊥ µ and ν − ⊥ µ.
We may now complete Part b) by showing [ν + ⊥ µ and ν − ⊥ µ] ⇒ ν ⊥ µ. Let us make explicit the
properties associated with ν + ⊥ µ and ν − ⊥ µ by replacing the roles of E, F (from the beginning
of this proof) with A1 , A2 for ν + ⊥ µ and B1 , B2 for ν − ⊥ µ. We first note that since A1 , B1
are both µ-null, so too is A1 ∪ B1 . This is true since by looking at the following representation:
A1 ∪ B1 ≡ A1 t (B1 \A1 ), and in noting B1 \A1 ⊂ B1 , ∀E 0 ⊂ A1 ∪ B1 , ∃A01 ⊂ A1 , B10 ⊂ B1 s.t.
E 0 = B10 t A1 and both B10 and A01 are µ-null. Furthermore, since A1 ∪ A2 = X = B1 ∪ B2 , we have
X\(A1 ∪ B1 ) = A2 ∩ B2 , which is both ν + -null and ν − -null since A2 ∩ B2 ⊂ A2 and A2 ∩ B2 ⊂ B2 .
Thus, by setting E = A1 ∪ B1 , and F = A2 ∩ B2 , we see that indeed ν ⊥ µ.
Pn n
b) |ν|(E) = sup 1 |ν(Ej )| n ∈ N, E1 , . . . , En are disjoint, and t1 Ej = E .
Proof.
a) For the first equality, ∀F ⊂ E we have:
ν(F ) = ν + (F ) − ν − (F ) ≤ ν + (F ) ≤ ν + (E)
And so ν + (E) ≥ sup{ν(F ) | E ∈ M, F ⊂ E}. To see the reverse inequality, if P, N are our
Hahn Decomposition of ν, we naturally have ν + (E) = ν(E ∩ P ), and since E ∩ P ⊂ E, ν + (E) ≤
sup{ν(F ) | E ∈ M, F ⊂ E}, and so:
ν + (E) = sup ν(F ) E ∈ M, F ⊂ E
For the second inequality, this follows very similarly. Explicitly, ∀F ⊂ E, we have:
−ν(F ) = ν − (F ) − ν + (F ) ≤ ν − (F ) ≤ ν − (E)
And so ν − (F ) ≥ sup{−ν(F ) | F ∈ M, F ⊂ E} = − inf{ν(F ) | F ∈ M, F ⊂ E}. For the reverse
inequality, since ν − (E) = −ν(E ∩ N ), and since E ∩ N ⊂ E, ν − (E) ≤ sup{−ν(F ) | F ∈ M, F ⊂
E} = − inf{ν(F ) | F ∈ M, F ⊂ E}. Combining our two inequalities, we see:
ν − (E) = − inf ν(F ) F ∈ M, F ⊂ E
31
And so:
n
X n
G
|ν|(E) ≤ sup |ν(Ej )| n ∈ N, E1 , . . . , En are disjoint, and Ej = E
1 1
And so:
n
X n
G
|ν|(E) ≥ sup |ν(Ej )| n ∈ N, E1 , . . . , En are disjoint, and Ej = E
1 1
dν
Proof. Let us begin by defining fj := dµjj for j = 1, 2. Thus, if A1 × A2 is measurable, by the definition
of product measure and Radon-Nikodym derivative, we have:
Z Z
ν1 × ν2 (A1 × A2 ) = ν1 (A1 )ν2 (A2 ) = f1 dµ1 f2 dµ2
ZA1 A2
Z
= f1 χA1 dµ1 f2 χA2 dµ2
Z Z
= f1 f2 χA1 χA2 dµ1 dµ2
Z Z
∗
= f1 f2 d(µ1 × µ2 )
A1 ×A2
32
∗
Where we have = by Tonelli’s Theorem. Therefore, on A1 × A2 measurable, (f1 f2 )(µ1 µ2 ) = ν1 ν2 ; and
thus we also have equality on the algebra of finite unions of A1 × A2 ’s. Furhermore, by the uniqueness
of the extension from premeasure to measure, (f1 f2 )(µ1 µ2 ) = ν1 ν2 on M1 ⊗ M2 . We thus immediately
have that if (µ1 × µ2 )(E) = 0 ⇒ (ν1 × ν2 )(E) = 0, and so ν1 × ν2 << µ1 × µ1 . Finally, since the
Radon-Nikodym derivative is unique, we have:
d(ν1 × ν2 ) dν1 dν2
(x1 , x2 ) = f1 (x1 )f2 (x2 ) = (x1 ) (x2 )
d(µ1 × µ2 ) dµ1 dµ2
Let X = [0, 1], M = B[0,1] , m = Lebesgue measure, and µ = counting measure on M, then:
a) m << µ but dm 6= f dµ for any f .
b) µ has no Lebesgue decomposition with respect to m.
Proof.
a) Firstly, if E ∈ M and µ(E) = 0, then it must be that E = , and so m(E) = m() = 0; I.e.,
m << µ. Suppose for the sake of contradiction that dm = f dµ, then ∀x ∈ [0, 1] and E = {x}, we
have: Z Z
0 = m(E) = f dµ = dm = m(E) = 0
E E
Thus we must have that f ≡ 0 on [0, 1]. However:
Z Z
1 = m([0, 1]) = f dµ = 0dµ = 0
[0,1] [0,1]
then g = g 0 ν-a.e. (In Probability Theory, g is call the conditional expectation of f on N.)
33
Proof. Let us begin by definingR the measure λ s.t. dλ = f dµ and its integration is restricted to E ∈ N;
I.e., ∀E R∈ N, we have λ(E) = E f dµ. To easily see that λ << ν, note that if ν(E) = 0 ⇒ µ(E) = 0 ⇒
λ(E) = E ddµ = 0. We thus have shown the necessary conditions for us to invoke The Lebesgue-Radon-
Nikodym theorem. Explicitly, the Radon-Nikodym derivative, g = dλ dν exists and is ν-integrable where
f dµ = dλ = gdν; I.e.; Z Z Z
f dµ = dλ = gdν ∀E ∈ N
E E E
Finally, if g 0 satisfies E f dµ = E g 0 dν, then naturally dλ = g 0 dν, and since the Radon-Nikodym deriva-
R R
Proof. Suppose that d|ν| = f dµ as in the definition of |ν|. Then if E ∈ M, then we will have:
∗ ∗
ν(E) + ν(E c ) = ν(X) = |ν|(X) = |ν|(E) + |ν|(E c ), where we have = by assumption
And so:
ν(E c ) − |ν|(E c ) = |ν|(E) − ν(E)
Taking the real part of the LHS, and using (Folland) Proposition 3.13a (|ν(E)| ≤ |ν|(E)), we see that:
= νr (E c ) − |ν(E c )|
q
= νr (E c ) − νr2 (E c ) + νi2 (E c )
p
≤ νr (E c ) − νr2 (E c ) = 0
And so combining the fact that Re(LHS) ≤ 0 ≤ Re(RHS), but obviously since Re(LHS) = Re(RHS),
we must have that:
Re(|ν|(E) − ν(E)) = 0 ⇒ |ν|(E) = νr (E)
But since, again by (Folland) Proposition 3.13a, we have |ν(E)| ≤ |ν|(E), we see that this must be true
⇐⇒ νi (E) = 0, and so:
34
3.7 Folland 3.21
Prove the following Proposition:
Proposition. 3.7:
dν
Then µ1 = µ2 = µ3 = |ν|. (First show that µ1 ≤ µ2 ≤ µ3 . To see that µ3 = |ν|, let f = d|ν| and
apply (Folland) Prop 3.13. To see that µ3 ≤ µ1 , approximate f by a simple function.)
∞
X X∞
|ν(Ej )| ≤ |ν|(Ej ) by (Folland) Prop. 3.13a
j=1 j=1
Z
= |ν|(E) = d|ν|
E
Z
= |f |2 d|ν| by (Folland) Prop. 3.13b
ZE
= f f d|ν|
ZE
= f dν by (Folland) Prop. 3.9a
ZE n Z o
≤ f dν ∈ f dν |f | ≤ 1
E E
If we are assuming |f | ≤ 1, then f −1 (D) = X, and so we will have X = ∪n1 f −1 (B (zj )) as well. By
defining Bj := f −1 (B (zj )), we now perform the standard “shuffle” to make a disjoint sequence {Aj }n1
35
j
out of {Bj }n1 , namely we let A1 = B1 , and Aj = BP n
j \ ∪i=1 Bi , so that X = t1 Aj . Now, in following the
n
hint, we explicitly define the simple function φ := 1 zj χAj .
Naturally |φ| ≤ 1 and |f (x) − φ(x)| < . Thus:
Z Z Z Z
µ3 (E) ≤ f dν + ≤
f dν −
φdν +
φdν +
E
ZE Z
E
Z E
≤ f dν − φdν + φdν + by the Reverse Triangle Inequality
ZE E
Z E
≤ |f − φ|dν + φdν +
E
Z Z E
≤ d|ν| + φdν +
E E
Z
≤ |ν|(E) + + φdν
E
So, by letting → 0, we have µ3 (E) ≤ | E φd|ν|. Now, let us define {Ej }n1 by Ej = Aj ∩ E so that
R
E = tn1 Ej ; thus:
n
n
Z Z X X Z
µ3 (E) ≤
φdν =
zj χAj χE dν =
zj dν
E j=1 j=1 Ej
n n
X X
=
ν(Ej ) ≤ |zj ||ν(Ej )|
j=1 j=1
n
X
{Ej }n1 E = tn1 Ej
≤ |ν(Ej )| ∈
j=1
And so µ3 ≤ µ1 . Thus since we were able to show µ1 ≤ µ2 ≤ |ν| ≤ µ3 ≤ µ1 , every inequality above is
actually an equality and in fact: µ1 = µ2 = µ3 = |ν|.
To see this, suppose that > 0. By the definition of continuity of f at x, we know that ∃δ > 0 s.t. if
||x − y|| < δ, I.e., y ∈ Bδ (x), we have |f (x) − f (y)| < . We therefore yield the following inequality for
0 < r < δ: Z Z
1 1
|f (y) − f (x)| ≤ dy = Ar () =
m(Br (x)) Br (x) m(Br (x)) Br (x)
Thus, since was arbitrary, we may conclude that our limit is indeed = 0; I.e., x is in the Lebesgue set
of f .
36
3.9 Folland 3.25
Prove the following Proposition:
Proposition. 3.9:
m(E ∩ Br (x))
DE (x) = lim
r&0 m(Br (x))
b) Find examples of E and x s.t. DE (x) is a given number α ∈ (0, 1), or such that DE (x) does
not exist.
Proof.
a) Let us begin by defining ν(A) := m(E ∩ A) ∀A ∈ BRn . Then, by construction we have ν << m and
dν
dm = χE . Furthermore, since {Br (x)}r>0 vacuously satisfies the requirements for a set to shrink
nicely to x ∈ Rn , we may make use of (Folland) Theorem 3.22. Explicitly, by 3.22, we have:
For the second example, we are looking for an E and an x s.t. DE (x) does not exist. Suppose for
this example we turn our thoughts to BR1 (so that Br (x) = (x − r, x + r)). Let us now set x = 0,
and define E as follows:
∞
G 1 1 1 1 1 1 1 1
E= , = , t , t , t ···
n=1
22n+1 22n 8 4 32 16 128 64
Our strategy henceforth will be to compose a countable subsequence, rk , s.t. rk & 0 and where
m(E∩B k (x))
limk→∞ m(Br r(x)) will be undefined, therefore also rendering DE (x) undefined. To do this, we
37
set rk = 21k . Naturally, we have m(Brk (x)) = 22k = 2k−1
1
. We now decompose (with a slight abuse
∞ ∞ 0 ∞
of notation) {rk }1 = {r2l }1 + {r2l+1 }1 , I.e. separate rk into two subsequences, one where k is
even, the other when k is odd. For the former (k is even), we have:
X 1 X ∞
1 1 1 X 1 1 2 1
m(E ∩ Brk (x) ) = m 2n+1 , 2n = = k
= k
=
2 2 2 2n+1 2 n=0 2 2n+1 2 3 3 · 2k−1
n≥l n≥l
m(E∩Brk (x))
And so limk→∞ Brk (x) is undefined, and therefore so too is DE (x) by our previous reasoning.
Proof. If K cpt ⊂ BRn , then λ(K), µ(K) < (λ + ν)(K) < ∞. Furthermore, suppose that E, F form the
singular decomposition of λ, µ; I.e., Rn = E t F and ∀F1 ⊂ F, F1 ∈ BRn , µ(F1 ) = 0, and similarly for E
w.r.t. λ.
Suppose now that A ∈ BRn . By definition of λ + µ’s regularity, we know that:
(λ + µ)(A) = inf (λ + µ)(Uopen ) | U ⊃ E
Therefore, ∀ = 2−k , k ∈ N, ∃ Uopen s.t. (λ + µ)(Uk ) < (λ + µ)(A) + . Thus, we may construct a
countable sequence of these such Uk ’s, namely {Uk }∞
1 , for which when letting k → ∞, we have:
By our set up of the singular decomposition of λ, µ we also note that we may express (λ + µ)(Uk ) =
(λ + µ)(Uk ∩ E) + (λ + µ)(Uk ∩ F ) = µ(Uk ∩ E) + λ(Uk ∩ F ), and similarly for A, namely: (λ + µ)(A) =
µ(A ∩ E) + λ(A ∩ F ). Furthermore, since µ, λ are positive measures, and by construction Uk ⊃ A ⇒
Uk ∩ E ⊃ A ∩ E (and similarly Uk ∩ F ⊃ A ∩ F ), we have µ(Uk ∩ E) ≥ µ(A ∩ E) and λ(Uk ∩ F ) ≥ λ(A ∩ F ).
By applying the last result twice, we can reach the following result:
(λ + µ)(Uk ) − (λ + µ)(A) = λ(Uk ∩ F ) + µ(Uk ∩ E) − λ(A ∩ F ) − µ(A ∩ E)
≥ λ(Uk ∩ F ) − λ(A ∩ F )
= λ(Uk ∩ F ) + λ(Uk ∩ E) −λ(A ∩ F ) − λ(A ∩ E)
| {z } | {z }
=0 =0
= λ(Uk ) − λ(A) ≥ 0
38
For which we already showed that the LHS has limit = 0, and thus taking limits on every equation in
the above reasoning shows that limk→∞ λ(Uk ) = λ(A). Furthermore, by the exact same steps but in
swapping λ ↔ µ, we see that limk→∞ µ(Uk ) = µ(A) as well. Therefore, the same approximation by open
sets from above for the definition of (λ + µ)’s regularity also works as an approximation by open sets
from above for all sets A ∈ BRn for λ and µ, hence we have arrived at the definition of λ and µ being
regular measures.
4 Chapter 5
4.1 Folland 5.1
Prove the following Proposition:
Proposition. 4.1:
If X is a normed vector space over K (= R or C), then addition and scalar multiplication are
continuous
from X × X and K × X to X. Moreover, the norm is continuous from X to [0, ∞); in
fact, ||x|| − ||y|| ≤ ||x − y||.
Proof. Let us define A : X × X → X as the addition map (I.e., defined as A(x, y) = x + y). The by
construction, A is a linear map from the NVS X × X to the NVS X. We thus will have (for (x, y) ∈ X × X):
||A(x, y)|| = ||x + y|| ≤ ||x|| + ||y|| ≤ 2 max{||x||, ||y||} = 2||(x, y)||
And therefore by (Folland) Proposition 5.2, since the above shows A is bounded, A must also be contin-
uous.
Let now define M : X × X → X as the scalar multiplication map (I.e., defined as M (α, x) = αx). Suppose
now that > 0 and we choose δ = min{1, }. Then if (α, x) ∈ K × X such that ||(α, x)|| < δ, we have:
And so is continuous at (0, 0), and hence again by (Folland) Proposition 5.2, M is continuous.
Lastly, let again > 0, but now set δ = . If x, y ∈ X such that ||x − y|| < δ, then:
||x|| = ||x − y + y|| ≤ ||x − y|| + ||y|| ⇒ ||x|| − ||y|| ≤ ||x − y||
And similarly for ||y|| − ||x|| ≤ ||y − x|| = ||x − y||. Therefore:
||x|| − ||y|| ≤ ||x − y|| < δ =
L(X, Y) is a vector space and the function || · || defined by (Folland, Equation 5.3) is a norm on it.
In particular, the three expressions on the right of (5.3) are always equal.
39
Proof. We begin by defining:
||T ||1 := sup ||T x|| | ||x|| = 1
||T x||
||T ||2 := sup | x 6= 0
||x||
||T ||3 := sup C | ||T x|| ≤ C||x|| ∀x ∈ X
As
in (Folland) Equation 5.3. We thus begin by showing ||·||1 = ||·||2 = ||·||3 . Firstly, if x ∈ X, x 6= 0, then
x/||x|| = 1 and so T (x)/||x|| = T (x/||x||) ≤ || · ||1 . Since this is true ∀x ∈ X, we may take the supremum
and hence ||T ||2 ≤ ||T ||1 . Next, again if x ∈ X and ||x|| = 1, then ||T x|| ≤ ||T ||3 , and again taking the
supremum implies ||T ||1 ≤ ||T ||3 . Lastly, again supposing x ∈ X, we simply have ||T x|| ≤ ||T ||2 , therefore
||T ||3 ≤ ||T ||2 . Summarizing we have: ||T ||1 ≤ ||T ||3 ≤ ||T ||2 ≤ ||T ||1 , and hence all our inequalities
above are actually equalities, and proving the equivalence of the above forms of (5.3).
To prove that || · || does indeed define a norm, suppose S, T ∈ L(X, Y), and x ∈ X. We thus have:
||(S + T )x|| = ||Sx + T x|| ≤ ||Sx|| + ||T x|| ≤ (||S|| + ||T ||)||x|| ⇒ ||S + T || ≤ ||S|| + ||T ||
And finally, ||T || = 0 ⇐⇒ ||T x|| = 0 ∀x ∈ X and T ≡ 0. Hence we’ve shown all the conditions for || · ||
to be a norm.
Proof. Let X be a subspace of X and X denote its closure. Firstly, by definition, 0 ∈ X. The other
property that we need to show is that if that if x, y ∈ X, and a, b ∈ K, then ax + by ∈ X as well. Since
x, y ∈ X, we know that ∃ {xj }∞ n
1 ⊂ X and {yj }1 ⊂ X s.t. xn → x and yn → y with respect to the norm,
|| · || on X. So, ∀/2 > 0, ∃N1 , N2 ∈ N s.t. ||xn − x|| < /2 and ||yn − y|| < /2 ∀n ≥ N1 , N2 respectively.
So, ∀n ≥ N = max(N1 , N2 ), we have:
(axn + byn ) − (ax + by) ≤ ||axn − ax|| + ||byn − by|| = |a|||xn − x|| + |b|||yn − y|| < 2 =
2
And so since axn + byn → ax + by, and axn + byn ∈ X it implies ax + by ∈ X by the definition of X.
Therefore, X is indeed a subspace of X.
40
Proposition. 4.4:
Suppose
Pn that XPis a finite-dimensional vector space. Let e1 , . . . , en be a basis for X and define
n
|| 1 aj ej ||1 = 1 |aj |.
a) || · ||1 is a norm on X.
Pn
b) The map (a1 , . . . , an ) → 1 aj ej is a continuous form K n with the usual Euclidean topology
to X with the topology defined by || · ||1 .
c) {x ∈ X | ||x||1 = 1} is compact in the topology defined by || · ||1 .
d) All norms on X are equivalent. (Compare any norm to || · ||1 .)
Proof.
Pn
a) We can first see that ||x||1 = 0 ⇐⇒ x = 0 since 1 |aj | = 0 ⇐⇒ aj = 0 ∀j = 1, . . . , n, and
0 := 0e1 + · · · + 0en .
n n
n n
X X X X
||x + y||1 =
αj ej + βj e j =
(αj + βj )ej = |αj + βj |
j=1 j=1 1 j=1 1 j=1
n
X n
X
≤ |αj | + |βj | = ||x||1 + ||y||1
j=1 j=1
And hence we have shown the three conditions for || · ||1 to be a norm on X.
Pn
From Part a), by dropping the absolute values in expressions of the form 1 |aj |, and replacing it by
b) P
n
Pn now becomes an equality, and hence the rest proves that T : K → X,
n
1 aj ej , the one inequality
where T (a1 , · · · , an ) = 1 aj ej , is a linear map. We may now invoke (Folland) Proposition 5.2,
which states T is continuous ⇐⇒ T is continuous at 0.
||x − 0|| = ||x|| = (a21 + · · · a2n )1/2 < δ ⇒ a2i ≤ (a21 + · · · + a2n ) < δ 2 ∀i = 1, . . . , n
n
X
||T x||1 =
aj ej = |a1 | + · · · + |an | < n
=
j=1 1 n
Pn
c) We begin by showing Γ := {(a1 , . . . , an ) ∈ K n | n
1 |aj | = 1} ⊂ K is compact. To see this, we
can simply show that Γ is closed and boundedP since Γ ⊂ K n = Cn or Rn . The boundness of Γ is
n
easy to see since: ||(a1 , . . . , an )||2 = ||x||2 := ( 1 a2j )1/2 ⇒ |aj | ≤ 1 ∀j = 1, . . . , n ⇒ B2 (0) ⊃ Γ,
hence Γ is bounded.
41
To see Γ is closed, we show that Γc is open. If x ∈ Γc , then:
n
X
x ∈ (a1 , . . . , an ) | |aj | =
6 1
1
n
X n
X
≡ (a1 , . . . , an ) | |aj | < 1 t (a1 , . . . , an ) | |aj | > 1 := Γ1 t Γ2
1 1
Pn Pn
= (x1 , . . . , xn ), then 1 |xj | < 1 or 1 |xj | > 1. Assume x ∈ Γ1 , and y ∈ K n . Letting
I.e., if x P
n
1 = 1 − 1 |xj | > 0, then in taking δ1 = 1 /n, we have:
1
||x − y||2 < δ1 ⇒ |xi − yi | ≤ ||x − y||2 < δ1 = ∀i ∈ {1, . . . , n}
n
Xn X n
⇒ ||x − y||1 = |xj − yj | < n =1− |xj |
j=1
n j=1
n
X n
X n
X
⇒ |xj | − |yj | < 1 − |xj | since |a| − |b| < |b − a|
j=1 j=1 j=1
Xn
⇒ |yj | < 1
j=1
Pn
And so B1 /n (x) ⊂ Γ1 , so Γ1 is open. Now suppose x ∈ Γ2 . Letting 2 = 1 |xj | − 1 and y ∈ K n
as before. Then letting δ2 = 2 /2, we have:
2
||x − y||2 < δ2 ⇒ |xi − yi | ≤ ||x − y||2 < δ2 = ∀i ∈ {1, . . . , n}
n
Xn X n
⇒ ||x − y||1 = |xj − yj | < n = |xj | − 1
j=1
n j=1
n
X n
X n
X
⇒ |xj | − |yj | < |xj | − 1 since |b| − |a| < |b − a|
j=1 j=1 j=1
Xn
⇒ |yj | > 1
j=1
And so B2 /n (x) ⊂ Γ2 , and hence Γ2 is open. Now since Γc = Γ1 t Γ2 , we can now conclude that
Γc is open, and hence Γ is closed, and hence compact. Furthermore, in Part b), we showed that T
(as defined in Part b) is continuous. Therefore, since:
T (Γ) = {x ∈ X | ||x||1 = 1}
We may now conclude that since Γ is compact, so too is {x ∈ X | ||x||1 = 1} in the topology defined
by || · ||1 .
d) Suppose || · || : χ → R≥0 is an arbitrary norm on X. We recall that to show || · || and || · ||1 are
equivalent, we need to find C1 , C2 > 0 s.t. C1 ||x||1 ≤ ||x|| ≤ C2 ||x||1 ∀x ∈ X. If x = 0, then
||x||1 = ||x|| since both are norms, selecting any C1 ≤ C2 where C1 , C2 > 0 proves the equivalence
of these norms for x = 0; therefore, assume x 6= 0.
Pn
If we let C2 = max({||ej ||n1 ), then if x ∈ X ⇒ x = 1 xj ej , then:
n n
∗ X X ∗
||x|| ≤ |xj |||ej || ≤ C2 = C2 ||x||1 where we have ≤ from the ∆-inequality
j=1 j=1
42
So we have found an appropriate C2 .
We now claim that || · || is continuous in the topology defined by || · ||1 . To see this, let > 0, and
δ = /n. If x, y ∈ X and ||x − y||1 < δ, then by what we found above:
||x − y|| ≤ C2 ||x − y||1 < C2 =
C2
Hence completing our proof since we found both C1 , C2 which satisfy the necessary inequality.
Let C k ([0, 1]) be the space of functions on [0, 1] possessing continuous derivatives up to order k
on [0, 1], including one-sided derivatives at the endpoints.
Proof.
a) We’ll proceed to prove this claim through induction. Suppose k = 0, then the forward case of
f ∈ C([0, 1]) implying f is differentiable on (0, 1) and and limx&n f (x) and limx%1 f (x) existing is
by the definition of C([0, 1]).
Now, for the backward direction (k = 0), suppose f ∈ C((0, 1)), limx&n f (x), and limx%1 f (x) exist
- this, however, is simply the definition of f ∈ C([0, 1]).
(j) (j) (j)
Let L0 := limx&n f (x) and L1 := limx%n f (j) (x). Now assume the property above holds for
k = n − 1. The forward direction is simply by definition. For the backward direction, if we wish to
(j)
show that f being k times differentiable on (0, 1) and limx&n f (x) and limx%n f (j) (x) existing for
j ≤ n implies f ∈ C k ([0, 1]), we may proceed as follows. Firstly, by the existence of the one sided
(j)
derivatives, we know that ∀ > 0, ∃δ > 0 such that if 0 < x < δ, then |f (j) (x) − L0 | < , ∀j ≤ n.
(j)
Furthermore, WLOG, we may omit the L1 case since all we need to chance in the argument is
43
that δ < x < 1 instead of 0 < x < δ. Moreover, by the mean value theorem, ∃x̂ ∈ (0, δ) s.t.
f (j−1) (x) − f (j−1) (0) = (x − 0)f (j) (x̂) = xf (j) (x̂) Therefore:
(j−1)
(x) − f (j−1) (x)
f
− L0 = |f (j) (x̂) − L1 | < since x̂ ∈ (0, δ)
x
And therefore by the fundamental theorem of calculus we may conclude that g = f 0 , and so fn → f
in C 1 .
We now make our inductive step. Assume the statement is true up until j = k. Suppose then that
(k+1) (k) (k+1)
{fn }∞
1 is Cauchy in C
k+1
and fn → f in C k and fn → g in C k+1 . Therefore, fn → fn
(k+1) (k+1) (k+1)
in C, and fn → g in C. We therefore may conclude that fn → fn in C, and ultimately
(k+1)
fn → f in C .
To finish our proof, we need to prove that || · || is indeed a norm. Firstly, if f 6≡ 0, then naturally
||f || =
6 0, so ||f || = 0 ⇐⇒ f ≡ 0. Since || · || is simply a sum of other norms, the triangle inequality
and absolutely scalability are both trivially immediate like definiteness.
5 Chapter 6
5.1 Folland 6.3
Prove the following Proposition:
Proposition. 5.1:
If 1 ≤ p < r ≤ ∞, Lp ∩ Lr is a Banach space with norm ||f || = ||f ||p + ||f ||r , and if p < q < r, the
inclusion map Lp ∩ Lr → Lq is continuous.
Proof. We begin by first showing that Lp ∩ Lr is a Banach Space w.r.t. ||f || = ||f ||p + ||f ||r (I.e., show
Lp ∩ Lr a normed vector space and complete w.r.t. ||f ||).
The fact that || · ||r and || · ||p are norms implies || · || is a norm. Firstly, || · || ≥ 0 since || · ||p , || · ||r ≥ 0.
Now, suppose f, g ∈ Lr ∩ Lp , and λ ∈ K, then we have:
||f + g|| = ||f + g||p + ||f + g||r ≤ ||f ||p + ||g||p + ||f ||r + ||g||r = ||f || + ||g||
44
||λf || = ||λf ||p + ||λf ||r = |λ| ||f ||p + |λ| ||f ||r = |λ| ||f ||
||f || = 0 ⇐⇒ ||f ||p = ||f ||r = 0 ⇐⇒ f ≡ 0 µ-a.e.
We can also immediately see that Lp ∩ Lr is a vector space since if u, v ∈ Lp ∩ Lr , then u, v ∈ Lp and
Lr , and so all our conditions for being a vector subspace are satisfied since both Lp and Lr are vector
subspaces.
Suppose now that {fn }∞ p r
1 be a Cauchy sequence in L ∩ L . By noting that ∀n, m ∈ N, we have
||fn − fm ||p ≤ ||fn − fm || and ||fn − fm ||r ≤ ||fn − fm ||, and hence {fn }∞ p r
1 are also Cauchy in L and L .
p r
We can thus define g and h as lim fn in L and L respectively. Let > 0, then ∃ N ∈ N s.t. if we take
δ = (p+1)/p , then letting ||fn − g||p < δ, and in setting E := {x ∈ X | ≤ |fn (x) − g(x)|}, we have:
Z Z Z
1 1 1 1 p 1 p
µ(E) = p p dµ ≤ p |fn − g|p dµ ≤ p |fn − g|p dµ = p ||fn − g||p < p δ =
E E
Suppose now that > 0 and f, g ∈ Lp ∩ Lr , then if ||f − g|| < δ = , we have ||f − g||q ≤ ||f − g|| < by
the above inequality. Hence ι : Lp ∩ Lr → Lq is uniformly continuous (and naturally continuous as well).
Proof. We begin by showing || · ||, as defined, is a norm. Firstly, || · || ≥ 0 since || · ||p , || · ||r ≥ 0. Now,
suppose f1 , f2 ∈ Lr + Lp , and λ ∈ K, then we have:
n o
||f1 + f2 || = inf ||g||p + ||h||r f1 + f2 = g + h
n o
= inf ||g1 + g2 ||p + ||h1 + h2 ||r f1 + f2 = g + h = (g1 + g2 ) + (h1 + h2 )
n o
≤ inf ||g1 ||p + ||g2 ||p + ||h1 ||r + ||h2 ||r f1 + f2 = g + h = (g1 + g2 ) + (h1 + h2 )
n o n o
≤ inf ||g1 ||p + ||h1 ||r f1 = g1 + h1 + inf ||g2 ||p + ||h2 ||r f2 = g2 + h2
= ||f1 || + ||f2 ||
45
n o
||λf || = inf ||λg||p + ||λh||r λf = λ(g + h)
n o
= inf |λ| ||g||p + |λ| ||h||r λf = λ(g + h)
n o
= |λ| inf ||g||p + ||h||r f = g + h
= |λ| ||f ||
||f || = 0 ⇐⇒ ||f ||p = ||f ||r = 0 ∀g, h s.t. f = g + h ⇐⇒ f ≡ 0 µ-a.e.
We can also immediately see that Lp + Lr is a vector space since if u, v ∈ Lp + Lr , then u = u1 + u2 , v =
v1 + v2 where u1 , v1 ∈ Lp and u2 , v2 ∈ Lr , and so all our conditions for being a vector subspace are
satisfied since both Lp and Lr are vector subspaces.
To show completeness, we make use of (Folland) Theorem 5.1 which states that a P normed vector space, X,
∞
is complete ⇐⇒ every absolutely convergent series in X converges. So, suppose 1 fn be an absolutely
convergent series in Lp + Lr . By the definition of inf and || · ||, we know that ∀n ∈ N, ∃ gn ∈ Lp , hn ∈ Lr
s.t. fn = gn P + hn where ||gn ||p + ||hn ||r < ||fn || + 2−n . Therefore,
P∞ from this
P∞ inequality, and since both
∞ ∞ −n
are absolutely convergent, so too will 1 gn and 1 hn . Since Lp and Lr are
P
1 fn and 1P 2
N PN
Banach spaces, 1 gn → g ∈ Lp and 1 hn → h ∈ Lr . Furthermore, P∞ byPdefinition || · || ≤ || · ||p and
∞
||·|| ≤ ||·||r , so combining these two reverse inequalities, we have 1 fn = 1 (gn +hn ), which therefore
p r p r
has a limit in L + L , explicitly g + h ∈ L + L . We have thus show all the necessary conditions for
Lp + Lr to be a Banach Space w.r.t. || · ||.
Suppose p < q < r and f ∈ Lq . Let E := {x ∈ X | 1 < |f (x)|}. Thus, by the construction of E, we
therefore have: |f χE |p ≤ |f χE |q and |f χE c |p ≤ |f χE c |q (I.e., f χE ∈ Lp , f χE ∈ Lr ), and hence:
||f || = ||f χE + f χE c || ≤ ||f χE ||p + ||f χE ||r ≤ ||f χE ||q + ||f χE c ||q = ||f ||q
Suppose now that > 0 and f, g ∈ Lq , then if ||f − g||q < δ = , we have ||f − g|| ≤ ||f − g||q < by the
above inequality. Hence ι : Lq → Lp + Lr is uniformly continuous (and naturally continuous as well).
Proof.
a) We first prove the following Lemma:
Lemma. 5.1: Chebyshev’s Inequality
p
||f ||p
µ(Et ) ≤
t
Where Et = {x ∈ X | |f (x)| ≥ t} and p ∈ (0, ∞).
46
Proof. Let g(x) = xp if x ≥ t, and 0 otherwise. We thus have 0 ≤ tp χEt ≤ |f |p χEt , and hence:
p
||f ||p
Z Z
1 p 1 p
µ(Et ) = p t χEt dµ ≤ p |f | dµ ≤
t t Et tp
Now back to the problem at hand. For the forward direction, we proceed via the contrapositive,
I.e., suppose ∃ > 0 s.t. ∀E ⊂ M(X), µ(E) 6∈ (0, ). From Chebyshev’s Inequality, we know that
||f ||p p
∃T s.t. ∀t ≥ T , µ(Et ) = 0 since µ(Et ) ≤ t → 0, and so |f | ≤ T a.e. So:
Z Z Z Z
|f |q dµ = |f |q dµ + |f |q dµ ≤ T q µ(Et ) + |f |p dµ < ∞
Et Etc Etc
And so f ∈ Lq .
For the converse, suppose ∀ > 0, ∃E ∈ M(X) s.t. µ(E) ∈ (0, ). Let us define {Fn }∞
1 where
0 < µ(Fn ) < 1/n so that µ(Fn ) → 0. By defining Gn := Fn \ ∪∞ n+1 Fm , we must have 0 <
µ(Fn ) ≤ µ(∪∞n Gm ). Furthermore, by taking subsequences, we may actually assume now that
0 < µ(Gm ) ≤ 2−m . Now if we define:
∞
X −1/q
f := µ(Gn ) χGn n−2/p (≥ 0)
n=1
Then we have:
∞ ∞
π2
Z Z Z X
−p/q X 1
|f |p dµ = f p dµ = µ(Gn ) χGn n−2 dµ = 2
= <∞
n=1 n=1
n 6
b) For the forward direction, the proof here is completely analogous to that in a). For the converse,
−1/q −1/(p+1)
by substituting µ(Gn ) for µ(Gn ) , and noting that now we have 2m ≤ µ(Gm ) < ∞
−m
instead of 0 < µ(Gm ) ≤ 2 , the same results as in a) still hold.
c) For the case of q = ∞, we have L∞ 6⊂ Lp ⇐⇒ µ(X) = ∞, since if |f |p < C ∈ R≥0 , we have:
Z Z
p
|f | dµ ≤ C dµ ≤ Cµ(X) < ∞
If f ∈ Lp ∩ L∞ for some p < ∞, so that f ∈ Lq ∀q > p, then ||f ||∞ = limq→∞ ||f ||q .
47
Proof. We may first assume f 6≡ 0 a.e. by the triviality of this case. From (the proof of Folland)
Proposition 6.10, we know that:
1−p/q p/q
||f ||q ≤ ||f ||∞ ||f ||p
And so: 1−p/q p/q
lim sup ||f ||q ≤ lim sup ||f ||∞ ||f ||p = ||f ||∞
q→∞ q→∞
Furthermore, by our initial assumption, we have ||f ||∞ > 0. Suppose now that 0 < a < ||f ||∞ and
Ea := {x ∈ X | |f (x)| ≥ a}. We thus have:
Z
p q 1/q q 1/q
aq µ(Ea ) ≤ ||f ||p ≤ |f |q dµ ≤ ||f ||q ⇒ aq µ(Ea ) ≤ ||f ||q
Ea
1/q
⇒ lim inf a µ(Ea ) ≤ lim inf ||f ||q
q→∞ q→∞
And so we must have all our inequalities become equalities: hence limq→∞ ||f ||q = ||f ||∞ .
Suppose 1 ≤ p < ∞. If fn , f ∈ Lp and fn → f a.e., then ||fn − f ||p → 0 ⇐⇒ ||fn ||p → ||f ||p .
[Use Exercise 20 in (Folland) 2.3.]
Proof. For the forward direction, if ||fn − f ||p → 0, by the triangle inequality we have:
||fn ||p − ||f ||p ≤ ||fn − f ||p → 0
And we therefore have ||fn ||p → ||f ||p .
For the converse, suppose ||fn ||p → ||f ||p . We now quickly prove the following result:
If z, w ∈ C, then |z − w|p ≤ 2p−1 (|z|p + |w|p ) ∀p ≥ 1
By the second derivative test, g(z) = |z|p is convex (I.e., g(tz + (1 − t)w) ≤ tg(z) + (1 − t)g(w)). So, if
we set t = 1/2, and move the 2p over to the other side, we have:
z − w p
p p−1 p p
|z − w| ≤ 2 (|z| + |w| ) ⇐⇒ ≤ 1 |z|p + 1 |w|p
2 2 2
For which the latter is recognizably true due to the convexity of | · |p for p ≥ 1 (and in making a change
of variables w0 = −w)
Carrying on, let us define gn := 2p−1 (|f |p + |fn |p ) − |f − fn |p . By the above inequality, we know that
gn ≥ 0, and so we may apply Fatou’s Lemma:
Z Z
p p
2 ||f ||p ≤ lim inf gn = 2 ||f ||p − lim sup |f − fn |p dµ
p p
n→∞ n→∞
48
5.6 Folland 6.14
Prove the following Proposition:
Proposition. 5.6:
If g ∈ L∞ , the operator T defined by T f = f g is bounded on Lp for 1 ≤ p ≤ ∞. Its operator
norm is at most ||g||∞ with equality if µ is semi-finite.
Proof. Firstly, we may assume g 6≡ 0 due to the triviality of this case. We now proceed to see that:
Z Z
p p p p
p p p
|f |p dµ = ||g|∞ | ||f ||p
||T f ||p = |f g| dµ = |f | |g| dµ ≤ ||g||∞ ≤ since |g| ≤ ||g||∞
⇒ ||T || ≤ ||g||∞
To see equality if µ is semi-finite, suppose 0 < < ||g||∞ , By µ’s semi-finitness, ∃ E s.t. ||g||∞ − <
|g| ∀x ∈ E. Thus, we have:
||T χE ||p = ||gχE || > ||g||∞ − ||χE ||p ⇒ ||T || > ||g||∞ − ⇒ ||T || ≥ ||g||∞
Where we have the last implication by ’s arbitrarily, and to satisfy both equalities, we must have
||g||∞ = ||T ||.
49