Chapter 5 Lp-spaces 5.1 Reduction to the Case `p and Lp The main (and only) result of this section is the following Theorem. Theorem 5.1.1. Let 1 p < 1 and let (⌦, ⌃, µ) be a separable measure space, i.e. ⌃ is generated by a countable set of subsets of ⌦. Then there is a countable set I so that Lp (⌦, ⌃, µ) is isometrically isomorphic to Lp [0, 1] p `p (I) or `p (I). Moreover, if (⌦, ⌃, µ) has no atoms we can choose I to be empty and, thus, Lp (⌦, ⌃, µ) is isometrically isomorphic to Lp [0, 1]. Proof. First note that the assumption that ⌃ is generated by a countable set say D ⇢ P(⌦)Simplies that Lp (µ) is separable. Indeed, the algebra generated by D is A = 1 n=1 An , where An is defined by recursively for every n 2 N as follows: A1 = D, and, assuming, An is defined we let first A0n+1 = k n[ j=1 Bj : k 2 N, Bj 2 An or Bjc 2 An o and then An+1 = k n\ j=1 o Bj : k 2 N, Bj 2 An or Bjc 2 An . This proves that A is countable. Then we observe that span(1A : A 2 A) is dense in Lp (µ) We first reduce to the -finite case. 117 118 CHAPTER 5. LP -SPACES Step 1: Lp (⌦, ⌃, µ) is isometrically isomorphic to a space Lp (⌦0 , ⌃0 , µ0 ) where (⌦0 , ⌃0 , µ0 ) is a -finite measure space. Let (fn ) ⇢ Lp (⌦, ⌃, µ) be a dense sequence in Lp (⌦, ⌃, µ) and define [ ⌦0 = |fn | > 0 . n2N Since {|fn | > 0} is a countable union of sets of finite measure, namely [ {|fn | > 0} = {|fn | > 1/m} m2N ⌦0 is also -finite. Moreover, for any f 2 Lp (⌦, ⌃, µ) it follows that {|f | > 0} ⇢ ⌦0 µ a.e. Therefore we can choose ⌃0 = ⌃|⌦0 = {A 2 ⌃ : A ⇢ ⌦0 } and µ0 = µ|⌃0 . Step 2: Assume (⌦, ⌃, µ) is a -finite measure space. Let I be the set of all atoms of (⌦, ⌃, µ). SinceSµ is -finite, I is countable, and µ(A) < 1 for all A 2 I. We put ⌦0 = ⌦ \ A2I A, ⌃0 = ⌃|⌦0 and µ0 = µ|⌃0 . Then ! ⇣Z ⌘ T : Lp (⌦, ⌃, µ) ! `p (I) p Lp (⌦0 , ⌃0 , µ0 ), f 7! f dµ : A 2 I , f |⌦0 , A is an isometry onto `p (I) p Lp (⌦0 , ⌃0 , µ0 ). Now either µ0 = 0 or it is an atomless -finite measure. In the next step we reduce to the case of µ being an atomless probability measure. Step 3: Assume that (⌦, ⌃, µ) is -finite, atomless and not 0. Then there is an atomless probability µ0 on (⌦, ⌃) so that Lp (⌦, ⌃, µ) is isometrically isomorphic to the space Lp (⌦, ⌃, µ0 ). Since (⌦, ⌃, µ) is -finite there is an f 2 L1 (⌦, ⌃, µ), with f (!) > 0 for all ! 2 ⌦ and kf k1 = 1. Let µ0 be the measure whose Radon Nikodym derivative with respect to µ is f (thus µ0 is a probability measure) and consider the operator T : Lp (⌦, ⌃, µ) ! Lp (⌦, ⌃, µ0 ), g 7! g · f 1/p , which is an isometry onto Lp (⌦, ⌃, µ0 ). Step 4: Reduction to [0, 1]. Assume (⌦, ⌃, µ) is an atomless countably generated probability space. Let (Bn ) ⇢ ⌃ be a sequence which generates ⌃. By induction we choose for each n 2 N0 a finite ⌃-partition Pn = (n) (n) (n) (P1 , P2 , . . . Pkn ) of ⌦ with the following properties: (5.1) {B1 , B2 , . . . Bn } ⇢ (Pn ) (the -algebra generated by Pn ), 5.1. REDUCTION TO THE CASE `P AND LP (5.2) (5.3) (n) µ(Pi ) 2 n 119 , for i = 1, 2 . . . kn , Pn is a subpartition of Pn each i 2 {1, . . . kn 1} 1 if n > 1, i.e. for there are sn (i) tn (i) in {1, . . . kn }, so that tn (i) (n 1) Pi [ = (n) Pj . j=sn (i) Put for n 2 N and 1 i kn h X ⌘ X (n) (n) (n) P̃i = µ(Pj ), µ(Pj ) , if j < kn and ji 1 (n) P̃kn = h X ji (n) µ(Pj ), jkn 1 (n) (n) X (n) µ(Pj ) jkn i (n) and P̃ (n) = (P̃1 , P̃2 , . . . , P̃kn ). Then P̃ (n) is a Borel partition of [0, 1], S (n) (n) with (P̃i ) = µ(Pi ), for each i kn , and n2N P̃ (n) generate the Borel -algebra on [0, 1]. For n 2 N put Vn = kn nX ai i=1 (n) Pi Then Vn is a vector space and V = Similarly Ṽ , with Ṽn = kn nX ai i=1 o : ai scalars , S n Vn is a dense subspace of LP (µ). o : ai scalars , (n) P̃i is a dense subspace of Lp [0, 1], and T : V ! Ṽ , kn X i=1 ai (n) Pi 7! kn X i=1 ai (n) P̃i , is an isometry whose image is dense in Lp [0, 1]. Thus T extends to an isometry from Lp (µ) onto Lp [0, 1]. Exercises 120 1. CHAPTER 5. LP -SPACES Prove that for 1 p 1, [ q>p Lq [0, 1] ( Lp [0, 1].