5.2. UNIFORM CONVEXITY AND SMOOTHNESS OF LP 5.2 121 Uniform Convexity and Uniform Smoothness of Lp , 1 < p < 1 Let (⌦, ⌃, µ) be a measure space. The first goal of this section is to prove the following Theorem 5.2.1. Let 1 < p < 1 and denote the modulus of uniform convexity of Lp (µ) by p . Then for any 1 < p < 1 there is a cp > 0 so that. ( cp "2 if 1 < p < 2 p (") cp "p if 2 < p < 1. Lemma 5.2.2. Assume ⇠, ⌘ 2 R a) If 2 p < 1, then b) If 0 < p 2 |⇠ + ⌘|p + |⇠ ⌘|p 2(|⇠|p + |⌘|p ). |⇠ + ⌘|p + |⇠ ⌘|p 2(|⇠|p + |⌘|p ). If p 6= 2 equality in (a) and (b) only holds if either ⇠ or ⌘ is zero. Proof. If p = 2 we have equality by the binomial formula. If 2 < p < 1 and ↵, 2 R, we apply Hölder’s inequality to the function {1, 2} ! {↵2 , 2 }, 1 7! ↵2 , 2 7! 2 , the counting measure on {1, 2}, and the exponents p/2 and p/(p ↵2 + (5.4) 2 (|↵|p + | |p )2/p 2(p |↵|p + | |p (↵2 + 2)/p 2 p/2 (2 p)/2 ) 2 2). , and, thus, . If 0 < p < 2 we can replace p by 4/p and obtain |↵|4/p + | |4/p (↵2 + 2 2/p (p 2)/p ) 2 , and if we replace |↵| and | | by |↵|p/2 and | |p/2 respectively, we obtain (5.5) |↵|2 + | |2 p p (|↵|p + | |p )2/p 2(p (2 p)/2 |↵| + | | 2 2 2)/p 2 p/2 (|↵| + | | ) , or . 122 CHAPTER 5. LP -SPACES Since 0 ⇠2 1 ⌘2 + ⇠2 we derive that |⇠|p (|⌘|2 + |⇠|2 )p/2 (5.6) ( ⇠2 ⌘ 2 +⇠ 2 ⇠2 ⌘ 2 +⇠ 2 if 2 < p if 2 > p. Forming similar inequalities by exchanging the roles of ⌘ and ⇠ and adding them we get ( (|⌘|p + |⇠|p )p/2 if 2 < p (5.7) |⌘|p + |⇠|p (|⌘|p + |⇠|p )p/2 if 2 > p. Note that equality in (5.7) can only hold if ⌘ = 0 or ⇠ = 0. Letting now ↵ = |⇠ + ⌘| and = |⇠ ⌘| we deduce from (5.4) and (5.7) if p > 2 |⇠ + ⌘|p + |⇠ ⌘|p |⇠ + ⌘|2 + |⇠ = 2 ⇠2 + ⌘ 2 p/2 ⌘|2 p/2 (2 p)/2 2 2 |⌘|p + |⇠|p , which finishes the proof of part (a), while part (b) follows from applying (5.5) and (5.7). Corollary 5.2.3. Let 0 < p < 1 and f, g 2 Lp (µ). Then ( 2(kf kpp + kgkpp ) if p 2 p p kf + gkp + kf gkp 2(kf kpp + kgkpp ) if p 2. If p 6= 2 equality only holds if f · g = 0 µ-almost everywhere. Lemma 5.2.4. Let 1 < p < 2. Then there is a positive constant C = C(p) so that ⇣ s t 2 s + t 2 ⌘1/2 ⇣ |s|p + |t|p ⌘1/p (5.8) + . C 2 2 Proof. We can assume without loss of generality that s = 1 > |t| and need therefore to show that for some C > 0 and all t 2 [ 1, 1] we have (5.9) ⇣1 C t ⌘2 (t) = ⇣ 1 + |t|p ⌘2/p 2 ⇣ 1 + t ⌘2 2 . 5.2. UNIFORM CONVEXITY AND SMOOTHNESS OF LP 123 Since is strictly positive on [ 1, 0] we only need to find C so that (5.8) holds for all t 2 [0, 1]. Since ⇠ 7! ⇠ 1/p is strictly concave it follows for all 0<t<1 ⇣ 1 tp ⌘2/p ⇣ 1 t ⌘2 + > + , 2 2 2 2 we only need to show that (5.10) lim t!1 (t) > 0. (1 t)2 We compute i d2 d h (2/p)+1 1 p (2/p) 1 p 1 (t) = 2 (1 + t ) t (1 + t) dt2 dt 2 = 2 (2/p)+1 (2 p)(1 + tp )(2/p) 2 t2p 2 1 + 2 (2/p)+1 (p 1)(1 + tp )(2/p) 1 tp 2 2 and thus d (t) dt d2 (t) dt2 t=1 = (2 t=1 = 0, and p)(1/2) + (p 1) (1/2) = (p 1)/2 > 0 Applying now twice the L’Hospital rule, we deduce our wanted inequality (5.10) Via integrating, Lemma 5.2.4 yields Corollary 5.2.5. If 1 < p 2 and f, g 2 Lp (µ) and if C = C(p) is as in Lemma 5.2.4, it follows (5.11) ⇣ f g C 2 + f + g 2 ⌘1/2 2 p ⇣ |f |p + |g|p ⌘1/p 2 =2 1/p ⇣ p kf kp + kgkp ⌘1/p . Proposition 5.2.6. If 1 p < q < 1 and fj 2 Lp , j = 1, 2 . . . then n X j=1 kfi kqp !1/q n ⇣X j=1 |fj |q ⌘1/q . p 124 CHAPTER 5. LP -SPACES Proof. We can assume without loss of generality that n X j=1 kfi kqp = 1. We estimate n ⇣X j=1 |fj | q ⌘1/q n X = j=1 p kfj kqp ⇣ |f | ⌘q j kfj kp !1/q = p !p/q !1/p ⇣ |f | ⌘q j kfj kp j=1 ! n ⇣ |f | ⌘p 1/p X j kfj kqp kfj kp n X kfj kqp j=1 p p (We use the concavity of the function ⇠ 7! ⇠ p/q .) n X j=1 kfj kqp ⇣ |f | ⌘p j kfj kp 1/p = 1, 1 which proves our claim. Proof of Theorem 5.2.1. For 2 p < 1 we will deduce our claim from Corollary 5.2.3. For f, g 2 Lp (µ), with kf k = kgk = 1, we deduce from the first inequality in Corollary 5.2.3 2p = 1⇥ k(f + g) 2 g)kp + k(f + g) + (f (f g)kp ] kf + gkp + kf and thus, using the approximation (2p +⇠)1/p = 2+ p1 21 that kf + gk 2p kf gkp )1/p = 2 1 1 2 p p kf p ⇠ +o(⇠), gkp + o(kf 2 p + f +g 2 2 ⌘1/2 p ⇣ f g 2 f +g 2 ⌘1/2 + C 2 we deduce gkp ), which implies our claim. Now assume that 1 < p < 2. Let f, g 2 SLp (µ) with " = kf Let C = C(p) be the constant in Corollary 5.2.5. We deduce from Proposition 5.2.6 and Corollary 5.2.5 that ⇣ f g C gkp p gkp > 0. 5.2. UNIFORM CONVEXITY AND SMOOTHNESS OF LP ⇣ |f |p + |g|p ⌘1/p =2 Solving for k(f + g)/2kp leads to r f +g f g 1 2 C p 2 1/p 2 p ⇣ p kf kp + kgkp =1 125 ⌘1/p = 1. "2 + o("2 ) 2C which implies our claim. Exercises 1. We say that a Banach space X is finitely representable in a Banach space Y , if for every finite dimensional subspace F of X and every " > 0 there is a finite dimensional subspace E of Y , so that dBM (E, F ) < 1 + ". Show that if Y is uniform convex and X is finitely representable in Y , then X is also uniformly convex. 2.⇤ We say that a Banach space X is crudely finitely representable in a Banach space Y , if there is a constant K so that for every finite dimensional subspace F of X there is a finite dimensional subspace E of Y , so that dBM (E, F ) < 1 + ". Show that if Y is uniform convex and X is finitely representable in Y , then there is an equivalent norm on X which turns X into a uniformly convex space.