Math 515 Professor Lieberman October 27, 2004 HOMEWORK #9 SOLUTIONS Chapter 6 15. By problem 14, k · k∞ is a norm on l∞ , so we only have to show that c and c0 are linear spaces. But, if hξn i and hηn i are elements of c and if α and β are real numbers, then lim αξn + βηn = α lim ξn + β lim ηn . It follows that hαξn + βηn i is also a convergent sequence, so it’s in c. Moreover, if both sequences have zero as a limit, then we conclude that hαξn + βηn i also has limit zero. To see that they are complete, suppose hξn i is a Cauchy sequence of elements of c (so ξn = hξn,m i). Again problem 14 shows that this sequence has a limit η = hηn i in l∞ . For simplicity of notation, we write Ln for the limit of the sequence ξn . We claim that hLn i is a Cauchy sequence, so it has a limit L and that L = lim ηn . To prove this claim, let ε > 0 be given. Then there is a constant N so that ε sup |ξn,m − ξj,m | < 3 m for all m, provided n, j ≥ N (because hξn i is Cauchy). Moreover, for n ≥ N , there is a constant M (n) such that ε |ξn,m − Ln | < 3 provided m ≥ M (n). Fixing n, j ≥ N , we take m ≥ max{M (n), M (j)}. Then |Ln − Lj | ≤ |Ln − ξn,m | + |ξn,m − ξj,m | + |ξj,m − Lj | < ε. It follows that hLn i is a Cauchy sequence, and we write L for its limit. Given ε > 0, there is an N so that ε sup |ξn,m − ηm | < 2 m if n ≥ N . There is also a constant J such that |Ln − L| < 4ε if n ≥ J and a constant M such that |ξn,m − Ln | < 4ε if m ≥ M and n = max{N, J}. For m ≥ M and n = max{N, J}, it follows that |ηm − L| ≤ |ηm − ξn,m | + |ξn,m − Ln | + |Ln − L| < ε, so L = lim ηm . It follows that c is complete. Moreover, if each ξn is in c0 , then lim ηm = 0, so c0 is also complete. 17. First, we assume that p > 1. Here is the proof for finite measure spaces. Without loss of generality kgkq > 0. Then, given ε, there is δ > 0 such that Z εq |g|q < (4M )q E for any set E with µE < δ. In addition, by problem 11.13b, the sequence converges in measure to f , so there is a positive integer N such that µEn < δ for n ≥ N , where ε En = x : |fn − f |(x) > . 2(µX)1/p kgkq 1 2 Writing Fn = X ∼ En , we have Z Z Z |fn − f ||g| + |fn − f ||g| = En Z q |fn − f ||g| Fn 1/q ε + 2(µX)1/p kgkq Z |g| ≤ kf − fn kp |g| Fn ε ε (µX)1/p kgkq = ε ≤ 2M + 2M 2(µX)p kgkq if n ≥ N . It follows that Z Z fn g → f g. If X has infinite measure, then problem 11.21a implies that Y = {x ∈ X : g(x) 6= 0} is σ-finite. Then Z Z Z Z fn g = fn g and = f g, X Y X Y so we may assume that X is σ-finite. Writing X = ∪Xj , with Xj ⊂ Xj+1 and µXj < ∞ for all j, we conclude that Z Z Z (fn − f )g = (fn − f )g + (fn − f )g. Xj X∼Xj Given ε > 0, there is an integer J such that Z ε q |g|q < 4M X∼Xj if j ≥ J and hence Z ε 2 X∼XJ for all n. The finite measure situation just proved shows that there is an integer N such that Z ε |fn − f ||g| < 2 Xj if n ≥ N . It follows that Z (fn − f )g < ε R R for n ≥ N and hence fn g → f g. The result is not true for p = 1, even on a finite measure space. Take µ to be Lebesgue measure on [0, 1], define fn by ( n if 0 ≤ x < n1 , fn = 0 if n1 ≤ x ≤ 1, R R and let g ≡ 1. Then fn → 0 a.e. but fn g = 1 and f g = 0. (fn − f )g ≤ 21. (a) If kgk = 0, take f = 1. Otherwise, take f = sgn g. (b) Set E(ε) = {x : |g(x)| ≥ kgk∞ − ε}, and take f = sgn gχE(ε) . Then Z Z fg = |g| ≥ (kgk∞ − ε)mE(ε) = (kgk∞ − ε)kf k1 . E(ε) 3 23. We first note that lp is just Lp (N, n), where N is the set of positive integers and n is counting measure. The proof of Proposition 80 shows that, if η ∈ l1 , then the equation ∞ X (1) G(ξ) = ξν ζn ν=1 ∞ l and defines a bounded linear functional on kGk = kηk1 . Hence this gives a bounded linear functional on c and on c0 . We first claim that every bounded linear functional F on c0 has this form. To prove this claim, write eν = hen,ν i for the vector in c0 defined by ( 1 if n = ν, eν,n = 0 if n 6= ν, and set ην = F (eν ). We now define G by (1), and we show that G = F . To show that G = F , we fix ξ ∈ l∞ , and we define a sequence hζn i of vectors in c0 by ( ξn if ν ≤ n, ζn,ν = 0 if ν > n. Since ξ ∈ c0 , it follows that ξν → 0 as ν → ∞ and hence ζn → ξ in l∞ . By linearity of F and G, we have G(ζn ) = F (ζn ) for all n and hence G(ξ) = F (ξ), so every bounded linear functional on c0 has the form described above. We now represent a bounded linear functional F on c as a sum. We define Ξ to be the vector with Ξν = 1 for all ν and set α = F (Ξ). We also write F0 for the functional F restricted to c0 . The claim now is that F (ξ) = F0 (ξ − LΞ) + αL with L = lim ξn . To prove this claim, we use linearity to infer that F (ξ) = F (ξ − LΞ) + F (LΞ). But ξ − LΞ ∈ c0 , so F (ξ − LΞ) = F0 (ξ − LΞ), and F (LΞ) = LF (Ξ) = αL. It follows from (i) that there is a vector η ∈ l1 such that X F (ξ) = ην [ξν − L] + αL. Conversely, given a vector η ∈ l1 and any α ∈ R, this equation gives a bounded linear functional on c, so this is our representation.