Chapter 6 Basic Inequalities and Lebesgue Spaces
6.2 The Lebesgue Spaces
6.2.1. The L p -Spaces
(ii) Assume thatg :C −→[0,∞) is a bounded, continuous, concave func- tion, and set ˆC ={(x, t)∈RN ×R : x∈C andt≤g(x)}. Show that ˆC is a closed, convex subset of RN+1. Next, define ˆF:E −→ Cˆ by ˆF = F
g◦F
, note that ˆFisµ-integrable, and apply the first part to see that itsµ-integral is an element of ˆC. Finally, notice that
Z Fˆdµ∈Cˆ =⇒ Z
g◦Fdµ≤g Z
Fdµ
. Exercise 6.1.10. Suppose that u∈C2 [0,1];R
satisfies u(0) = 0 =u(1).
The goal of this exercise is to show that
(∗) u(t) =−
Z
[0,1]
(s∧t−st)u00(s)ds fort∈[0,1].
In particular, ifu00≤0, thenu≥0.
(i) Use integration by parts to show that u(t) =tu0(0) +
Z
[0,t]
(t−s)u00(s)ds fort∈[0,1].
(ii) Using (i), show thatu0(0) =−R
[0,1](1−s)u00(s)dsand therefore that (∗) holds.
6 Basic Inequalities and Lebesgue Spaces
the collection of equivalence classes [f]∼µ ofR-valued, measurable functions f satisfyingkfkLp(µ;R)<∞. Once again, I will consistently abuse notation by using f to denote its own equivalence class [f]∼µ.
Obviously kαfkLp(µ;R) = |α|kfkLp(µ;R) for all p ∈ [1,∞), α∈ R, and B- measurable f’s. Also, by (3.1.10) and Minkowski’s inequality, we have the triangle inequality
kf1+f2kLp(µ;R)≤ kf1kLp(µ;R)+kf2kLp(µ;R)
for all p∈ [1,∞) and f1, f2 ∈Lp(µ;R). Moreover, it is a simple matter to check that these relations hold equally well when p=∞. Thus, each of the spacesLp(µ;R) is a vector space. In addition, because of our convention and Markov’s inequality (Theorem 3.1.6),kfkLp(µ;R)= 0 if and only iff = 0 as an element ofLp(µ;R). Hence,kf2−f1kLp(µ;R)determines a metric onLp(µ;R), and I will write fn −→ f in Lp(µ;R) when {fn : n≥ 1} ∪ {f} ⊆Lp(µ;R) andkfn−fkLp(µ;R)−→0.
The following theorem simply summarizes obvious applications of the re- sults in §§3.1 and 3.2 to the present context. The reader should verify that each of the assertions here follows from the relevant result there.
Theorem 6.2.1. Let(E,B, µ)be a measure space. Then, for anyp∈[1,∞]
andf, g∈Lp(µ;R),
kgkLp(µ;R)− kfkLp(µ;R)
≤ kg−fkLp(µ;R).
Next suppose that {fn : n≥1} ⊆Lp(µ;R)for some p∈[1,∞]and thatf is anR-valued measurable function on(E,B).
(i)Ifp∈[1,∞) andfn −→f in Lp(µ;R), thenfn −→f in µ-measure. If fn−→f in L∞(µ;R), then fn−→f uniformly off of a set ofµ-measure 0.
(ii)Ifp∈[1,∞]andfn−→f in µ-measure or(a.e.,µ), then kfkLp(µ;R)≤ limn→∞kfnkLp(µ;R). Moreover, if p ∈ [1,∞) and, in addition, there is a g ∈ Lp(µ;R) such that|fn| ≤ g (a.e.,µ)for each n∈Z+, then fn −→f in Lp(µ;R).
(iii)If p∈[1,∞]andlimm→∞supn≥mkfn−fmkLp(µ;R)= 0, then there is an f ∈Lp(µ;R) such that fn −→ f in Lp(µ;R). In other words, the space Lp(µ;R)is complete with respect to the metric determined byk · kLp(µ;R).
Finally, we have the following variants of Theorem 3.2.14 and Corollary 3.2.15.
(iv)Assume thatµ(E)<∞ and thatp, q∈[1,∞). Referring to Theorem 3.2.14, define S as in that theorem. Then, for eachf ∈Lp(µ;R)∩Lq(µ;R), there is a sequence{ϕn: n≥1} ⊆ S such thatϕn−→f both inLp(µ;R)and in Lq(µ;R). In particular, if µ isσ-finite andB is generated by a countable collection C, then each of the spacesLp(µ;R),p∈[1,∞), is separable.
(v) Let (E, ρ) be a metric space, and suppose that µ is a measure on (E,BE)for which there exists a non-decreasing sequence of open setsEn %E satisfying µ(En)<∞ for each n≥1. Then, for each pair p, q ∈[1,∞)and f ∈ Lp(µ;R)∩Lq(µ;R), there is a sequence {ϕn : n ≥ 1} of bounded, ρ- uniformly continuous functions such thatϕn≡0off ofEnandϕn−→f both in Lp(µ;R)and inLq(µ;R).
The version of Lieb’s variation on Fatou’s Lemma forLp-spaces withp6= 1 is not so easy as the assertions in Theorem 6.2.1. To prove it we will need the following lemma.
Lemma 6.2.2. Letp∈(1,∞), and suppose that {fn : n≥1} ⊆Lp(µ;R) satisfies supn≥1kfnkLp(µ;R) <∞ and that fn −→ 0 either in µ-measure or (a.e.,µ). Then, for everyg∈Lp(µ;R),
n→∞lim Z
|fn|p−1|g|dµ= 0 = lim
n→∞
Z
|fn| |g|p−1dµ.
Proof: Without loss of generality, we will assume that all of thefn’s as well as gare non-negative. Givenδ >0, we have that
Z
fnp−1g dµ= Z
{fn≤δg}
fnp−1g dµ+ Z
{fn>δg}
fnp−1g dµ
≤δp−1kgkpLp(µ;R)+ Z
{fn≥δ2}
fnp−1g dµ+ Z
{g≤δ}
fnp−1g dµ.
Applying H¨older’s inequality to each of the last two terms, we obtain Z
fnp−1g dµ≤δp−1kgkpLp(µ;R)
+kfnkp−1Lp(µ;R)
Z
{fn≥δ2}
gpdµ
!1p +
Z
{g≤δ}
gpdµ
!1p
. Since, by Lebesgue’s Dominated Convergence Theorem, the first term in the final brackets tends to 0 asn→0, we conclude that
n→∞lim Z
fnp−1g dµ≤δp−1kgkpLp(µ;
R)+ sup
n≥1
kfnkp−1Lp(µ;
R)k1{g≤δ}gkLp(µ;R)
for every δ > 0. Thus, after another application of Lebesgue’s Dominated Convergence Theorem, we get the first equality upon lettingδ&0.
To derive the other equality, apply the preceding with fnp−1 and gp−1 re- placing, respectively,fn andg and withp0= p−1p in place ofp.
6 Basic Inequalities and Lebesgue Spaces
Theorem 6.2.3 (Lieb). Let(E,B, µ)be a measure space,p∈[1,∞), and {fn : n≥1} ∪ {f} ⊆Lp(µ;R). If supn≥1kfnkLp(µ;R)<∞ andfn −→f in µ-measure or(a.e.,µ), then
n→∞lim Z
|fn|p− |f|p− |fn−f|p dµ= 0;
and thereforekfn−fkLp(µ;R)−→0if kfnkLp(µ;R)−→ kfkLp(µ;R).
Proof: The case p= 1 is covered by Theorems 3.2.3 and 3.2.5, and so we will assume thatp∈(1,∞). Given such ap, we will first check that there is a Kp<∞such that
(∗)
|b|p− |a|p− |b−a|p
≤Kp |b−a|p−1|a|+|a|p−1|b−a|
, a, b∈R. Since it is clear that (∗) holds for all a, b∈ R if it does for all a ∈ R\ {0}
and b∈R, we can assume thata6= 0 and divide both sides by |a|p, thereby showing that (∗) is equivalent to
|c|p−1− |c−1|p
≤Kp |c−1|p−1+|c−1|
, c∈R.
Finally, the existence of a Kp < ∞ for which this inequality holds can be easily verified with elementary consideration of what happens when cis near 1 and when|c|is near infinity.
Applying (∗) witha=fn(x) andb=f(x), we see that
|fn|p− |f|p− |fn−f|p
≤Kp |fn−f|p−1|f|+|fn−f||f|p−1 pointwise. Thus, by Lemma 6.2.2 withfnandgthere replaced by, respectively, fn−f andf here, our result follows.
Before applying H¨older’s inequality to the Lp-spaces, it makes sense to complete the definition of the H¨older conjugatep0that was given in Theorem 6.1.5 only for p∈(1,∞). Namely, I will take the H¨older conjugate of 1 to be
∞and that of ∞to be 1. Notice that this is completely consistent with the equation 1p+p10 = 1 used before.
Theorem 6.2.4. Let(E,B, µ)be a measure space.
(i)Iff andgare measurable functions on(E,B), then for everyp∈[1,∞], kf gkL1(µ;R)≤ kfkLp(µ;R)kgkLp0(µ;R).
In particular, iff ∈Lp(µ;R)andg∈Lp0(µ;R), thenf g∈L1(µ;R).
(ii)Ifp∈[1,∞)andf ∈Lp(µ;R), then kfkLp(µ;R)= sup
kf gkL1(µ;R):g∈Lp0(µ;R) andkgkLp0
(µ;R)≤1 .
In fact, if kfkLp(µ;R)>0, then the supremum on the right is achieved by the function
g= |f|p−1 kfkp−1Lp(µ;R)
.
(iii)More generally, for anyf that is measurable on (E,B), kfkLp(µ;R)≥sup
kf gkL1(µ;R):g∈Lp0(µ;R) andkgkLp0(µ;R)≤1 , and equality holds ifp= 1orp∈(1,∞)and eitherµ(|f| ≥δ)<∞for every δ >0orµisσ-finite.
(iv)Iff :E −→Ris measurable andµ(|f| ≥R)∈(0,∞)for someR≥0, then
kfkL∞(µ;R)= sup
kf gkL1(µ;R):g∈L1(µ;R) andkgkL1(µ;R)≤1 . Proof: Part (i) is an immediate consequence of H¨older’s inequality when p ∈ (1,∞). At the same time, when p ∈ {1,∞}, the conclusion is clear without any further comment. Given (i), (ii) as well as the inequality in (iii) are obvious.
Whenp= 1, equality in (iii) is trivial, since one can takeg=1. Moreover, in view of (ii), the proof of equality in (iii) forp∈(1,∞) reduces to showing that, under either one of the stated conditions, kfkLp(µ;R)=∞implies that the supremum on right-hand side is infinite. To this end, first suppose that µ(|f| ≥δ)<∞for every δ >0. Then, for eachn≥1, the function given by
ψn≡ |f|p−1
1[n1,n]◦ |f|
+n1{∞}◦f
is an element ofLp0(µ;R). Moreover, ifkfkLp(µ;R)=∞, then, by the Mono- tone Convergence Theorem,kψnkLp0(µ;R)−→ ∞. Thus, sincekf ψnkL1(µ;R)≥ kψnkp0
Lp0(µ;R),we see that
kf gnkL1(µ;R)−→ ∞ ifkfkLp(µ;R)=∞andgn≡ ψn 1 +kψnkLp0
(µ;R)
. To handle the case µ isσ-finite andµ(|f| ≥δ) =∞for some δ >0, choose {En: n≥1} ⊆ Bsuch that En%E andµ(En)<∞for every n≥1. Then it is easy to see that limn→∞kf gnkL1(µ;R)=∞when
gn≡ 1Γn
1 +µ(Γn)1−1p with Γn=En∩ {|f| ≥δ}.
SincekgnkLp0
(µ;R)≤1, this completes the proof of (iii).
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