Chapter 6 Basic Inequalities and Lebesgue Spaces
6.2 The Lebesgue Spaces
6.2.2. Mixed Lebesgue Spaces
6 Basic Inequalities and Lebesgue Spaces
Finally, to check (iv), first note that the right side dominates the left. To get the opposite inequality, define gM = µ(|f|≥M1[M,∞]◦f) for M ∈[0,∞) satisfying µ(|f| ≥ M)∈ (0,∞). Obviously, kgMkL1(µ;R) = 1 and kf gMkL1(µ;R) ≥M. If R = kfkL∞(µ;R), take M = R to get kf gRkL1(µ;R) ≥ kfkL∞(µ;R). If R <
kfkL∞(µ;R), get the same conclusion by considering M ∈
R,kfkL∞(µ;R)
. Thus, the left side dominates the right.
§6.2.2. Mixed Lebesgue Spaces: For reasons that will become clearer
E2, fn(·, x2) −→ f(·, x2) (a.e.,µ1), |fn(·, x2)| ≤ g(·, x2) (a.e.,µ1), and g(·, x2) ∈ Lp1(µ1;R). Thus, by part (ii) of Theorem 6.2.1, for µ2-almost every x2∈E2,kfn(·, x2)−f(·, x2)kLp1(µ1;R)−→0. In addition,
kfn(·, x2)−f(·, x2)kLp1(µ1;R)≤2kg(·, x2)kLp1(µ1;R)
forµ2-almost everyx2∈E2 and, by (∗) withg replacingf,
kg(·1,·2)kLp1(µ1;R)
Lp2(µ2;R)<∞.
Hence the required result follows after a second application of (ii) in Theorem 6.2.1.
We turn now to the final part of the lemma, in which both the measures µ1 andµ2 are assumed to be finite. In fact, without loss of generality, we will assume that they are probability measures. In addition, by the preceding, it is clear that, for eachf ∈L(p1,p2) (µ1, µ2);R
,
kf −fnkL(p1,p2)((µ1,µ2);R)−→0 where fn≡f1[−n,n]◦f.
Thus, we need only consider f’s that are bounded. Finally, becauseµ1×µ2
is also a probability measure, Jensen’s inequality and (∗) imply that kf−ψkL(p1,p2 )(µ1,µ2)≤ kf −ψkLq(µ1×µ2) where q=p1∨p2. Hence, all that remains is to show that, for every bounded,B1×B2-measurable f :E1×E2−→Rand >0, there is aψ∈ Gfor whichkf−ψkLq(µ1×µ2)< . But, by part (iv) of Theorem 6.2.1, the class of simple functions having the form
ψ=
n
X
m=1
am1Γ1,m×Γ2,m
with Γi,m ∈ Bi is dense in Lq(µ1×µ2;R). Thus, we will be done once we check that such a ψis an element of G. To this end, set I = {0,1}n
and, forη∈ I, define Γ1,η =Tn
m=1Γ1,m(ηm) where Γ(0)≡Γ{ and Γ(1)≡Γ. Then ψ(x1, x2) =
n
X
m=1
am
X
η∈I
ηm1Γ1,η(x1)
1Γ2,m(x2) =X
η∈I
1Γ1,η(x1)ϕη(x2),
where
ϕη =
n
X
m=1
ηmam1Γ2,m.
Since the Γ1,η’s are mutually disjoint, this completes the proof.
We can now prove the following continuous version of Minkowski’s inequality.
6 Basic Inequalities and Lebesgue Spaces
Theorem 6.2.7. Let (Ei,Bi, µi), i ∈ {1,2}, be σ-finite measure spaces.
Then, for any measurable functionf on(E1×E2,B1× B2),
kfkL(p1,p2 )((µ1,µ2);R)≤ kfkL(p2,p1 )((µ2,µ1);R) if1≤p1≤p2<∞.
Proof: Since it is easy to reduce the general case to the one in which both µ1 andµ2 are finite, we will take them to be probability measures from the outset.
Let G be the class described in the last part of Lemma 6.2.5. Given ψ = Pn
11Γ1,m(·1)ϕm(·2) from G, note that, since the Γ1,m’s are mutually dis- joint, |Pn
1am1Γ1,m|r=Pn
1|am|r1Γ1,m for anyr∈[0,∞) anda1, . . . , an∈R. Hence, by Minkowski’s inequality withp= pp2
1, kψkL(p1,p2)((µ1,µ2);R)=
"
Z
E2
n X
m=1
µ1(Γ1,m)|ϕm(x2)|p1
p2 p1
µ2(dx2)
#p12
=
n
X
m=1
µ1(Γ1,m)|ϕm(·2)|p1
1 p1
L
p2 p1(µ2;R)
≤
" n X
m=1
µ1(Γ1,m)
|ϕm|p1
L
p2 p1(µ2;R)
#p11
=
" n X
1
µ1(Γ1,m)kϕmkpL1p2(µ2;R)
#p11
=
Z
E1
n
X
m=1
1Γ1,m(x1)kϕmkpL1p2(µ2;R)µ1(dx1)
1 p1
=
Z
E1 n
X
m=1
1Γ1,m(x1)kϕmkpL2p2(µ2;R)
!
p1 p2
µ1(dx1)
1 p1
=
Z
E1
Z
E2
n
X
1
1Γ1,m(x1)ϕm(x2)
p2
µ2(dx2) pp12
µ1(dx1)
1 p1
=kψkL(p2,p1 )((µ2,µ1);R).
Therefore we are done when the functionf is an element ofG.
To complete the proof, letf be a measurable function on (E1×E2,B1×B2).
Clearly we may assume thatkfkL(p2,p1 )((µ2,µ1);R)<∞. Using the last part of Lemma 6.2.5, choose{ψn: n≥1} ⊆ Gsuch thatkψn−fkL(p2,p1 )((µ2,µ1);R)−→
0. Then, by Jensen’s inequality, one has thatkψn−fkL1((µ1×µ2);R)−→0, and therefore thatψn−→f inµ1×µ2-measure. Hence, without loss of generality, assume that ψn −→f (a.e.,µ1×µ2). In particular, by Fatou’s Lemma and Exercise 4.1.9, this means that
Z
E1
|f(x1, x2)|p1µ1(dx1)≤ lim
n→∞
Z
E1
|ψn(x1, x2)|p1µ1(dx1)
for µ2-almost every x2 ∈ E2; and so, by the result for elements of G and another application of Fatou’s Lemma, the required result follows forf.
Exercises for §6.2
Exercise 6.2.8. Let (E,B, µ) be a measure space and f : E −→ R a B- measurable function.
(i) Ifµis finite, show thatkfkLp(µ;R)≤µ(E)p1−1qkfkLq(µ;R)for 1≤p < q≤
∞. In particular, whenµis a probability measure, this means thatkfkLp(µ;R)
is non-decreasing as a function ofp.
(ii) WhenE is countable,B=P(E), andµis thecounting measure on E (i.e., µ({x}) = 1 for eachx∈E), show thatkfkLp(µ;R) is a non-increasing function ofp∈[1,∞].
Hint: First reduce to the case whenE={1, . . . , n}for somen∈Z+. Second, show that it suffices to prove that Pn
m=1apm ≤ 1 for every p ∈ [1,∞) and {am: 1≤m≤n} ⊆[0,1] withP
m=1am= 1, and apply elementary calculus to check this.
(iii) Ifµis finite orf isµ-integrable, show that, asp→ ∞,kfkLp(µ;R)−→
kfkL∞(µ;R) for anyB-measurablef :E −→R.
(iv) Let (E1,B1) and (E2,B2) be a pair of measurable spaces, and letµ2be aσ-finite measure on (E2,B2). Iff :E1×E2−→RisB1×B2-measurable, use (iii) to show thatx1∈E1 7−→ kf(x1,·)kL∞(µ2;R)∈[0,∞] isB1-measurable.
Hence, we could have defined L(p1,p2) (µ1, µ2);R
for allp1, p2∈[1,∞].
Exercise 6.2.9. Let a measure space (E,B, µ) and 1 ≤ q0 ≤ q1 ≤ ∞ be given. Iff ∈Lq0(µ;R)∩Lq1(µ;R), show that for everyt∈(0,1)
kfkLqt(µ;R)≤ kfktLq0(µ;R)kfk1−tLq1(µ;R) where 1 qt
= t q0
+1−t q1
. Note that this can be summarized by saying that p −logkfkLp(µ;R) is a concave function of 1p.
Exercise 6.2.10. If (E1,B1, µ1) and (E2,B2, µ2) are a pair of σ-finite mea- sure spaces andp∈[1,∞), show that the set of functions that can be written in the form Pn
m=1f1,m(x1)f2,m(x2), where n ≥ 1, {f1,m : 1 ≤ m ≤ n} ⊆ Lp(µ1;R), and{f2,m: 1≤m≤n} ⊆Lp(µ2;R), is dense inLp(µ1×µ2;R).
Hint: First reduce to the case in whichµ1 andµ2 are finite, and then apply part (iv) of Theorem 6.2.1 to handle this case.
Exercise 6.2.11. Let (E,B, µ) be a measure space,ga non-negative element ofLp(µ;R) for somep∈(1,∞), andf a non-negative,B-measurable function for which there exists aC∈(0,∞) such that
(∗) µ(f ≥t)≤ C
t Z
{f≥t}
g dµ, t∈(0,∞).
6 Basic Inequalities and Lebesgue Spaces
The purpose of this exercise is to show that (∗) allows one to estimate theLp- norm off in terms of that ofg whenp >1. The result here is a very special case of a general result known as Marcinkiewicz’s Interpolation Theorem.
(i) Setν(Γ) =R
Γg dµfor Γ∈ B, note that (∗) is equivalent to µ(f > t)≤ C
tν(f > t), t∈(0,∞), and use (5.1.7) or Exercise 5.1.11 to justify
kfkpLp(µ;R)=p Z
(0,∞)
tp−1µ(f > t)dt
≤Cp Z
(0,∞)
tp−2ν(f > t)dt= Cp p−1
Z
fp−1dν.
Finally, note that R
fp−1dν =R
fp−1g dµ, and apply H¨older’s inequality to conclude that
(∗∗) kfkpLp(µ;R)≤ Cp
p−1kfkp−1Lp(µ;R)kgkLp(µ;R).
(ii) Under the condition thatkfkLp(µ;R)<∞, it is clear that (∗∗) implies (6.2.12) kfkLp(µ;R)≤ Cp
p−1kgkLp(µ;R).
Now suppose that µ(E) <∞. After checking that (∗) forf implies (∗) for fR≡f ∧R, conclude that (6.2.12) holds first withfR replacingf and then, afterR% ∞, forf itself. In other words, whenµis finite, (∗) always implies (6.2.12).
(iii) Even ifµis not finite, show that (∗) implies (6.2.12) if, for every >0, µ(f > )<∞.
Hint: Given >0, considerµ=µB
{f > }
, note that (∗) withµimplies itself with µ, and use (ii) to conclude that (6.2.12) holds withµ in place of µ. Finally, let&0.
Exercise 6.2.13. In §3.3.2 I derived the Hardy–Littlewood inequality (3.3.7) and pointed out the one cannot pass from such an estimate to an estimate on the L1-norm. Nonetheless, use (3.3.7) together with Exercise 6.2.11 to show that
kM fkLp(λR;R)≤ 2p
p−1kfkLp(λR;R) for allp∈(1,∞).
§6.3 Some Elementary Transformations on Lebesgue Spaces