Exercises for Chapter 3
4.2 Definition and Elementary Properties of L p Spaces
Theorem 4.4 (Tonelli).Let F (x, y) : 1×2 → R be a measurable function satisfying
(a)
2|F (x, y)|dμ2<∞fora.e.x ∈1
and (b)
1
dμ1
2
|F (x, y)|dμ2<∞.
ThenF ∈L1(1×2).
Theorem 4.5 (Fubini).Assume thatF ∈ L1(1×2). Then for a.e. x ∈ 1, F (x, y) ∈ L1y(2)and
2F (x, y)dμ2 ∈ L1x(1). Similarly, for a.e. y ∈ 2, F (x, y)∈L1x(1)and
1F (x, y)dμ1∈L1y(2).
Moreover, one has
1
dμ1
2
F (x, y)dμ2=
2
dμ2
1
F (x, y)dμ1=
1×2
F (x, y)dμ1dμ2.
4.2 Definition and Elementary Properties of L
pSpaces
Definition. Letp∈Rwith 1< p <∞; we set Lp()=
f :→R;f is measurable and|f|p∈L1()
with
fLp = fp= -
|f (x)|pdμ .1/p
. We shall check later on that pis a norm.
Definition. We set L∞()=
f :→R
f is measurable and there is a constant C such that|f (x)| ≤Ca.e. on
with
fL∞ = f∞=inf{C; |f (x)| ≤Ca.e. on}. The following remark implies that ∞is a norm:
Remark1.Iff ∈L∞then we have
|f (x)| ≤ f∞ a.e. on.
Indeed, there exists a sequenceCnsuch thatCn→ f∞and for eachn,|f (x)| ≤ Cn a.e. on. Therefore|f (x)| ≤ Cn for allx ∈ \En, with |En| = 0. We set
E= ∪∞n=1En, so that|E| =0 and
|f (x)| ≤Cn ∀n, ∀x∈\E; it follows that|f (x)| ≤ f∞ ∀x∈\E.
Notation. Let 1≤p≤ ∞; we denote byptheconjugate exponent, 1
p + 1 p =1.
•Theorem 4.6 (Hölder’s inequality). Assume that f ∈ Lp and g ∈ Lp with 1≤p≤ ∞. Thenf g∈L1and
(1)
|f g| ≤ fpgp.
Proof. The conclusion is obvious ifp = 1 orp = ∞; therefore we assume that 1< p <∞. We recallYoung’s inequality:1
(2) ab≤ 1
pap+ 1
pbp ∀a≥0, ∀b≥0.
Inequality (2) is a straightforward consequence of the concavity of the function log on(0,∞):
log 1
pap+ 1 pbp
≥ 1
plog ap+ 1
plog bp =log ab.
We have
|f (x)g(x)| ≤ 1
p|f (x)|p+ 1
p|g(x)|p a.e.x ∈. It follows thatf g∈L1and
(3)
|f g| ≤ 1 pfp
p+ 1 pgp
p. Replacingf byλf (λ >0)in (3), yields
(4)
|f g| ≤ λp−1 p fp
p+ 1 λpgp
p.
Choosing λ = f−p1gpp/p (so as to minimize the right-hand side in (4)), we obtain (1).
1It is sometimes convenient to use the formab≤εap+CεbpwithCε=ε−1/(p−1).
4.2 Definition and Elementary Properties ofLpSpaces 93 Remark2.It is useful to keep in mind the following extension of Hölder’s inequality:
Assume thatf1, f2, . . . , fkare functions such that fi ∈Lpi, 1≤i≤kwith 1
p = 1 p1+ 1
p2 + · · · + 1 pk ≤1.
Then the productf =f1f2· · ·fk belongs toLpand fp≤ f1p1f2p2· · · fkpk.
In particular, iff ∈Lp∩Lqwith 1≤p≤q≤ ∞, thenf ∈Lrfor allr,p≤r≤q, and the following “interpolation inequality” holds:
fr ≤ fαpf1−α
q , where1 r =α
p +1−α
q , 0≤α≤1; see Exercise 4.4.
Theorem 4.7.Lpis a vector space and pis a norm for anyp,1≤p≤ ∞.
Proof. The casesp=1 andp = ∞are clear. Therefore we assume 1< p < ∞ and letf, g∈Lp. We have
|f (x)+g(x)|p≤(|f (x)| + |g(x)|)p≤2p(|f (x)|p+ |g(x)|p).
Consequently,f +g∈Lp. On the other hand, f +gpp =
|f +g|p−1|f +g| ≤
|f +g|p−1|f| +
|f +g|p−1|g|.
But|f+g|p−1∈Lp, and by Hölder’s inequality we obtain f +gpp≤ f +gpp−1(fp+ gp), i.e.,f +gp≤ fp+ gp.
•Theorem 4.8 (Fischer–Riesz).Lpis a Banach space for anyp,1≤p≤ ∞.
Proof. We distinguish the casesp= ∞and 1≤p <∞.
Case 1:p= ∞.Let(fn)be a Cauchy sequence isL∞. Given an integerk≥1 there is an integerNk such thatfm−fn∞≤ 1k form, n ≥Nk. Hence there is a null setEksuch that
(5) |fm(x)−fn(x)| ≤ 1
k ∀x ∈\Ek, ∀m, n≥Nk. Then we letE=
kEk—so thatEis a null set—and we see that for allx ∈\E, the sequencefn(x)is Cauchy (inR). Thusfn(x)→f (x)for allx ∈\E. Passing to the limit in (5) asm→ ∞we obtain
|f (x)−fn(x)| ≤ 1
k for allx∈\E, ∀n≥Nk.
We conclude thatf ∈ L∞andf −fn∞ ≤ 1k ∀n ≥ Nk; thereforefn → f inL∞.
Case 2: 1≤p <∞.Let(fn)be a Cauchy sequence inLp. In order to conclude, it suffices to show that a subsequence converges inLp.
We extract a subsequence(fnk)such that fnk+1−fnkp≤ 1
2k ∀k≥1.
[One proceeds as follows: choosen1such thatfm−fnp ≤ 12 ∀m, n ≥ n1; then choosen2≥n1such thatfm−fnp ≤ 212 ∀m, n≥n2etc.] We claim that fnk converges inLp. In order to simplify the notation we writefkinstead offnk, so that we have
(6) fk+1−fkp ≤ 1
2k ∀k≥1.
Let
gn(x)= n k=1
|fk+1(x)−fk(x)|, so that
gnp≤1.
As a consequence of the monotone convergence theorem,gn(x)tends to a finite limit, sayg(x), a.e. on, withg∈Lp. On the other hand, form≥n≥2 we have
|fm(x)−fn(x)| ≤ |fm(x)−fm−1(x)|+· · ·+|fn+1(x)−fn(x)| ≤g(x)−gn−1(x).
It follows that a.e. on, fn(x)is Cauchy and converges to a finite limit, sayf (x).
We have a.e. on,
(7) |f (x)−fn(x)| ≤g(x) forn≥2,
and in particular f ∈ Lp. Finally, we conclude by dominated convergence that fn−fp→0, since|fn(x)−f (x)|p →0 a.e. and also|fn−f|p ≤gp ∈L1. Theorem 4.9.Let(fn)be a sequence inLpand letf ∈Lpbe such thatfn−fp
→0.
Then, there exist a subsequence(fnk)and a functionh∈Lpsuch that (a)fnk(x)→f (x)a.e.on,
(b)|fnk(x)| ≤h(x) ∀k,a.e.on.
Proof. The conclusion is obvious whenp= ∞. Thus we assume 1≤p <∞. Since (fn)is a Cauchy sequence we may go back to the proof of Theorem 4.8 and consider
4.3 Reflexivity. Separability. Dual ofLp 95 a subsequence(fnk)—denoted by(fk)—satisfying (6), such thatfk(x)tends a.e. to a limit2f(x)withf ∈Lp. Moreover, by (7), we have|f(x)−fk(x)| ≤ g(x)
∀k, a.e. onwithg ∈Lp. By dominated convergence we know thatfk → f in Lp and thusf =fa.e. In addition, we also have|fk(x)| ≤ |f(x)| +g(x), and the conclusion follows.
4.3 Reflexivity. Separability. Dual of L
pWe shall consider separately the following three cases:
(A) 1< p <∞, (B) p=1, (C) p= ∞.
A. Study ofLp()for 1< p <∞.
This case is the most “favorable”:Lpis reflexive, separable, and the dual ofLp isLp.
•Theorem 4.10.Lpis reflexive for anyp, 1< p <∞.
The proof consists of three steps:
Step1 (Clarkson’s first inequality). Let 2≤p <∞.We claim that
(8)
f +g 2
p
p
+ f −g
2 p
p
≤ 1
2(fpp+ gpp) ∀f, g∈Lp. Proof of(8).Clearly, it suffices to show that
a+b 2
p+ a−b
2
p≤ 1
2(|a|p+ |b|p) ∀a, b∈R. First we note that
αp+βp ≤(α2+β2)p/2 ∀α, β ≥0 (by homogeneity, assumeβ =1 and observe that the function
(x2+1)p/2−xp−1
increases on[0,∞)). Choosingα= |a+2b|andβ = |a−2b|, we obtain a+b
2 p+
a−b 2
p≤
a+b 2
2+ a−b
2 2
p/2
= a2
2 +b2 2
p/2
≤1
2(|a|p+|b|p)
2A priori one should distinguishfandf: by assumptionfn→finLp, and on the other hand, fnk(x)→f(x)a.e.
(the last inequality follows from the convexity of the function x → |x|p/2 since p≥2).
Step2:Lp is uniformly convex, and thus reflexive for2 ≤ p < ∞. Indeed, let ε >0 and letf, g∈Lpwithfp≤1,gp≤1, andf−gp > ε. We deduce from (8) that
f +g 2
p
p
<1−/ε 2
0p
and thusf+2gp < 1−δ withδ = 1− [1−(ε2)p]1/p > 0. Therefore, Lp is uniformly convex and thus reflexive by Theorem 3.31.
Step3:Lpis reflexive for1< p≤2.
Proof. Let 1< p <∞. Consider the operatorT :Lp→(Lp)defined as follows:
Letu∈Lpbe fixed; the mappingf ∈Lp →
uf is a continuous linear functional onLp and thus it defines an element, sayT u, in(Lp)such that
T u, f =
u f ∀f ∈Lp. We claim that
(9) T u(Lp) = up ∀u∈Lp.
Indeed, by Hölder’s inequality, we have
|T u, f| ≤ up fp ∀f ∈Lp and thereforeT u(Lp) ≤ up.
On the other hand, set
f0(x)= |u(x)|p−2u(x) (f0(x)=0 ifu(x)=0).
Clearly we have
f0∈Lp, f0p =up−1
p and T u, f0 = upp; thus
(10) T u(Lp) ≥ T u, f0 f0p
= up.
Hence, we have shown thatT is an isometry fromLpinto(Lp), which implies that T (Lp)is a closed subspace of(Lp)(becauseLpis a Banach space).
Assume now 1< p≤2. SinceLpis reflexive (by Step 2), it follows that(Lp)
4.3 Reflexivity. Separability. Dual ofLp 97 is also reflexive (Corollary 3.21). We conclude, by Proposition 3.20, thatT (Lp)is reflexive, and as a consequence,Lpis also reflexive.
Remark3.In fact,Lpis also uniformly convex for1< p≤2. This is a consequence ofClarkson’s second inequality, which holds for 1< p≤2:
f +g 2
p
p
+ f −g
2 p
p
≤ 1
2fp
p+1 2gp
p
1/(p−1)
∀f, g∈Lp.
This inequality is trickier to prove than Clarkson’s first inequality (see, e.g., Prob- lem 20 or E. Hewitt–K. Stromberg [1]). Clearly, it implies thatLpis uniformly convex when 1< p≤2; for another approach, see also C. Morawetz [1] (Exercise 4.12) or J. Diestel [1].
•Theorem 4.11 (Riesz representation theorem).Let1 < p < ∞and letφ ∈ (Lp). Then there exists a unique functionu∈Lp such that
φ, f =
uf ∀f ∈Lp. Moreover,
up =φ
(Lp).
Remark4.Theorem 4.11 is very important. It says that every continuous linear func- tional onLpwith 1< p <∞can be represented “concretely” as an integral. The mapping φ → u, which is a linear surjective isometry, allows us to identify the
“abstract” space(Lp)withLp.
In what follows, we shall systematically make the identification (Lp)=Lp.
Proof. We consider the operatorT : Lp → (Lp) defined byT u, f = uf
∀u ∈ Lp,∀f ∈ Lp. The argument used in the proof of Theorem 4.10 (Step 3) shows that
T u(Lp) = up ∀u∈Lp.
We claim thatT is surjective. Indeed, letE=T (Lp). SinceEis a closed subspace, it suffices to prove thatE is dense in(Lp). Let h ∈ (Lp)satisfyh, T u = 0
∀u∈Lp. SinceLpis reflexive,h∈Lp, and satisfies
uh=0∀u∈Lp. Choosing u= |h|p−2h, we see thath=0.
Theorem 4.12.The spaceCc(RN)is dense inLp(RN)for anyp,1≤p <∞. Before proving Theorem 4.12, we introduce some notation.
Notation. Thetruncation operationTn:R→Ris defined by
Tnr=
⎧⎨
⎩
r if|r| ≤n, nr
|r| if|r|> n.
Given a setE⊂, we define thecharacteristic function3χE to be χE(x)=
1 ifx∈E, 0 ifx∈\E.
Proof. First, we claim that given f ∈ Lp(RN)andε > 0 there exist a function g∈L∞(RN)and a compact setKinRNsuch thatg=0 outsideKand
(11) f−gp < ε.
Indeed, letχn be the characteristic function ofB(0, n)and letfn = χnTnf. By dominated convergence we see that fn −fp → 0 and thus we may choose g = fn withn large enough. Next, givenδ > 0 there exists (by Theorem 4.3) a functiong1∈Cc(RN)such that
g−g11< δ.
We may always assume thatg1∞≤ g∞; otherwise, we replaceg1byTng1with n= g∞. Finally, we have
g−g1p≤ g−g11/p1 g−g11∞−(1/p)≤δ1/p(2g∞)1−(1/p). We conclude by choosingδ >0 small enough that
δ1/p(2g∞)1−(1/p)< ε.
Definition. The measure spaceis calledseparableif there is a countable family (En)of members ofMsuch that theσ-algebra generated by(En)coincides with M(i.e.,Mis the smallestσ-algebra containing all theEn’s).
Example.The measure space=RNis separable. Indeed, we may choose for(En) any countable family of open sets such that every open set inRNcan be written as a union ofEn’s. More generally, ifis aseparable metric spaceandMconsists of the Borel sets (i.e.,Mis theσ-algebra generated by the open sets in), thenis a separable measure space.
Theorem 4.13.Assume thatis a separable measure space. ThenLp()is sepa- rable for anyp,1≤p <∞.
We shall consider only the case = RN, since the general case is somewhat tricky. Note that as a consequence,Lp() is also separable for any measurable set⊂RN. Indeed, there is a canonical isometry fromLp()intoLp(RN)(the
3Not to be confused with theindicator functionIEintroduced in Chapter 1.
4.3 Reflexivity. Separability. Dual ofLp 99 extension by 0 outside); thereforeLp()may be identified with a subspace of Lp(RN)and henceLp()is separable (by Proposition 3.25).
Proof of Theorem4.13when= RN.LetRdenote the countable family of sets inRN of the form R = 1N
k=1(ak, bk)withak, bk ∈ Q. LetE denote the vector space overQgenerated by the functions(χR)R∈R, that is,Econsists of finite linear combinations with rational coefficients of functionsχR, so thatEis countable.
We claim that E is dense inLp(RN). Indeed, givenf ∈ Lp(RN)andε > 0, there exists somef1∈Cc(RN)such thatf −f1p < ε. LetR ∈Rbe any cube containing suppf1(the support off1). Givenδ >0 it is easy to construct a function f2 ∈ E such thatf1−f2∞ < δ andf2 vanishes outsideR: it suffices to split R into small cubes ofRwhere the oscillation (i.e., sup−inf) off1is less thanδ.
Therefore we havef1−f2p ≤ f1−f2∞|R|1/p < δ|R|1/p. We conclude that f −f2p<2ε, providedδ >0 is chosen so thatδ|R|1/p< ε.
B. Study ofL1().
We start with a description of the dual space ofL1().
•Theorem 4.14 (Riesz representation theorem).Letφ∈(L1). Then there exists a unique functionu∈L∞such that
φ, f =
uf ∀f ∈L1. Moreover,
u∞= φ(L1).
•Remark5.Theorem 4.14 asserts that every continuous linear functional onL1can be represented “concretely” as an integral. The mappingφ →u, which is a linear surjective isometry, allows us to identify the “abstract” space(L1)withL∞.In what follows, we shall systematically make the identification
(L1)=L∞.
Proof. Let(n)be a sequence of measurable sets insuch that= ∪∞n=1nand
|n|<∞ ∀n. Setχn=χn.
Theuniquenessofuis obvious. Indeed, supposeu∈L∞satisfies
uf =0 ∀f ∈L1.
Choosingf =χnsignu(throughout this book, we use the convention that sign 0= 0), we see thatu=0 a.e. onnand thusu=0 a.e. on.
We now prove the existence of u. First, we construct a function θ ∈ L2() such that
θ (x)≥εn >0 ∀x ∈n.
It is clear that such a functionθ exists. Indeed, we defineθto beα1on1, α2 on2\1, . . . , αnonn\n−1, etc., and we adjust the constantsαn>0 in such a way thatθ∈L2.
The mappingf ∈L2()→ φ, θfis a continuous linear functional onL2().
By Theorem 4.11 (applied withp=2) there exists a functionv∈L2()such that
(12) φ, θf =
vf ∀f ∈L2().
Setu(x)=v(x)/θ (x). Clearly,uis well defined sinceθ >0 on; moreover,uis measurable anduχn ∈L2(). We claim thatuhas all the required properties. We have
(13) φ, χng =
uχng ∀g∈L∞() ∀n.
Indeed, it suffices to choosef = χng/θ in (12) (note thatf ∈ L2()sincef is bounded onnandf =0 outsiden).
Next, we claim thatu∈L∞()and that
(14) u∞≤ φ(L1).
Fix any constantC >φ(L1) and set
A= {x∈; |u(x)|> C}.
Let us verify that A is a null set. Indeed, by choosingg=χAsignuin (13) we obtain
A∩n
|u| ≤ φ(L1)|A∩n| and therefore
C|A∩n| ≤ φ(L1)|A∩n|.
It follows that|A∩n| =0 ∀n, and thus A is a null set. This concludes the proof of (14).
Finally, we claim that
(15) φ, h =
uh ∀h∈L1().
Indeed, it suffices to chooseg =Tnh(truncation ofh) in (13) and to observe that χnTnh→hinL1().
In order to complete the proof of Theorem 4.14 it remains only to check that u∞= φ(L1). We have, by (15),
|φ, h| ≤ u∞h1 ∀h∈L1(),
4.3 Reflexivity. Separability. Dual ofLp 101
and thereforeφ(L1) ≤ u∞. We conclude with the help of (14).
•Remark6.The spaceL1()isnever reflexiveexcept in the trivial case where consists of a finite number of atoms—and thenL1()is finite-dimensional. Indeed suppose, by contradiction, thatL1()is reflexive and consider two cases:
(i)∀ε >0∃ω⊂measurable with 0< μ(ω) < ε.
(ii)∃ε >0 such thatμ(ω)≥εfor every measurable setω⊂withμ(ω) >0.
In Case (i) there is a decreasing sequence(ωn)of measurable sets such that μ(ωn) > 0 ∀n andμ(ωn) → 0 [choose first any sequence(ωk)such that 0 <
μ(ωk) <1/2kand then setωn=∞
k=nωk].
Let χn = χωn and define un = χn/χn1. Since un1 = 1 there is a subsequence—still denoted by un—and someu ∈ L1 such that un u in the weak topologyσ (L1, L∞)(by Theorem 3.18), i.e.,
(16)
unφ→
uφ ∀φ∈L∞. On the other hand, for fixedj, and n > j we have
unχj = 1. At the limit, as n→ ∞, we obtain
uχj =1∀j. Finally, we note (by dominated convergence) that uχj →0 asj → ∞—a contradiction.
In Case (ii) the spaceis purely atomic and consists of a countable union of distinct atoms (an)(unless there is only a finite number of atoms!). In that case L1()is isomorphic to1and it suffices to prove that1is not reflexive. Consider the canonical basis:
en=(0,0, . . . , 1
(n),0,0. . . ).
Assuming1is reflexive, there exist a subsequence(enk)and somex∈1such that enk xin the weak topologyσ (1, ∞), i.e.,
ϕ, enk −→
k→∞ϕ, x ∀ϕ ∈∞. Choosing
ϕ=ϕj =(0,0, . . . , 1
(j ),1,1, . . . )
we find thatϕj, x = 1 ∀j. On the other handϕj, x → 0 asj → ∞ (since x ∈1)—a contradiction.
C. Study ofL∞.
We already know (Theorem 4.14) that L∞ = (L1). Being a dual space,L∞ enjoys some nice properties. In particular, we have the following:
(i) The closed unit ball BL∞ is compact in the weak topology σ (L∞, L1) (by Theorem 3.16).
(ii) Ifis a measurable subset inRNand(fn)is a bounded sequence inL∞(), there exists a subsequence(fnk)and somef ∈ L∞()such thatfnk f in
the weak topology σ (L∞, L1)(this is a consequence of Corollary 3.30 and Theorem 4.13).
HoweverL∞()isnot reflexive, except in the trivial case whereconsists of a finite number of atoms; otherwiseL1()would be reflexive (by Corollary 3.21) and we know thatL1is not reflexive (Remark 6). As a consequence, it follows that thedual space(L∞)ofL∞containsL1(sinceL∞=(L1))and(L∞)isstrictly biggerthanL1. In other words, there are continuous linear functionalsφ onL∞ whichcannotbe represented as
φ, f =
uf ∀f ∈L∞and someu∈L1.
In fact, let us describe a “concrete” example of such a functional. Letφ0:Cc(RN)→ Rbe defined by
φ0(f )=f (0)forf ∈Cc(RN).
Clearlyφ0is a continuous linear functional onCc(RN)for the ∞norm. By Hahn–
Banach, we may extendφ0 into a continuous linear functionalφonL∞(RN)and we have
(17) φ, f =f (0) ∀f ∈Cc(RN).
Let us verify that there existsnofunctionu∈L1(RN)such that
(18) φ, f =
uf ∀f ∈L∞(RN).
Assume, by contradiction, that such a functionuexists. We deduce from (17) and
(18) that
uf =0 ∀f ∈Cc(RN)andf (0)=0.
Applying Corollary 4.24 (with=RN\{0})we see thatu=0 a.e. onRN\{0}and thusu=0 a.e. onRN. We conclude (by (18)) that
φ, f =0 ∀f ∈L∞(RN), which contradicts (17).
Remark7.The dual space ofL∞does not coincide withL1but we may still ask the question: what does(L∞) look like? For this purpose it is convenient to view L∞(;C) as a commutative C-algebra (see, e.g., W. Rudin [1]). By Gelfand’s theoremL∞(;C)is isomorphic and isometric to the spaceC(K;C)of continuous complex-valued functions on some compact topological spaceK(Kis the spectrum of the algebraL∞;Kis not metrizable except whenconsists of a finite number of atoms). Therefore(L∞(;C)) may be identified with the space of complex- valuedRadon measuresonKandL∞(;R)may be identified with the space of
4.3 Reflexivity. Separability. Dual ofLp 103 real-valued Radon measures onK; for more details, see Comment 3 at the end of this chapter, W. Rudin [1] and K. Yosida [1] (p. 118).
Remark8.The spaceL∞()isnot separableexcept when consists of a finite number of atoms. In order to prove this fact it is convenient to use the following.
Lemma 4.2.LetEbe a Banach space. Assume that there exists a family(Oi)i∈I such that
(i)for eachi∈I,Oi is a nonempty open subset ofE, (ii)Oi∩Oj = ∅ifi =j ,
(iii)I isuncountable.
ThenEisnotseparable.
Proof of Lemma4.2.Suppose, by contradiction, thatEis separable. Let (un)n∈N denote a dense countable set inE. For eachi∈I, the setOi∩(un)n∈N = ∅and we may choosen(i)such thatun(i) ∈Oi. The mappingi→n(i)is injective; indeed, if n(i) = n(j ), thenun(i) = un(j ) ∈ Oi ∩Oj and thus i = j. Therefore,I is countable—a contradiction.
We now establish that L∞()is not separable. We claim that there is an un- countable family(ωi)i∈I of measurable sets inwhich are all distinct, that is, the symmetric differenceωi ωj has positive measure fori =j. We then conclude by applying Lemma 4.2 to the family(Oi)i∈I defined by
Oi = {f ∈L∞(); f −χωi∞<1/2}
(note thatχω−χω∞=1 ifωandωare distinct). The existence of an uncountable family(ωi)is clear whenis an open set inRNsince we may consider all the balls B(x0, r)withx0∈andr >0 small enough.
Whenis a general measure space we splitinto its atomic parta and its nonatomic (=diffuse) partd; then we distinguish two cases:
(i) dis not a null set.
(ii) dis a null set.
In Case (i), then for each real numbert, 0< t < μ(d), there is a measurable setωwithμ(ω)=t; see, e.g., P. Halmos [1], A. J. Weir [1], or J. Neveu [1]. In this way, we obtain an uncountable family of distinct measurable sets.
In Case (ii) consists of a countable union of distinct atoms(an)(unless consists of a finite number of atoms). For any collection of integers,A⊂N, we define ωA=
n∈Aan. Clearly,(ωA)is an uncountable family of distinct measurable sets.
The following table summarizes the main properties of the spaceLp()when is a measurable subset ofRN:
Reflexive Separable Dual space Lpwith 1< p <∞ YES YES Lp
L1 NO YES L∞
L∞ NO NO Strictly bigger thanL1