Math 516 Professor Lieberman February 14, 2005 HOMEWORK #4 SOLUTIONS Chapter 10 38. (a) If xn → x weakly, then f (xn ) → f (x) for any f ∈ X ∗ . It follows that {ϕ[xn ]} is a family of bounded linear operators from the Banach space X ∗ to the normed linear space R such that, for each f ∈ X ∗ , there is a number Mf with |ϕ[xn ](f )| ≤ Mf for all xn . (In other words, a convergent sequence is bounded.) The uniform boundedness principle then implies that there is a number M such that kϕ[xn ]k ≤ M for all n. But ϕ is an isometry, so kxn k ≤ M for all n. (b) =⇒: From part (a) hkxn ki is bounded. In addition, for each integer n, the function fn , defined by fn (hξi i) = ξn is clearly linear and p 1/p |ξn | = (|ξn | ) ≤ ∞ X !1/p p |ξi | = khξi ik. i=1 It follows that fn is a bounded linear function on lp and hence ξn,m = fn (xm ) → fn (x) = ξm . ⇐=: From Fatou’s Lemma, we know that kxk ≤ M , where M is any number such that hkxn ki ≤ M . Next, we fix a bounded linear functional f on lp . By the Riesz Representation P Theorem, there is a sequence η ∈ lq (q = p/(p − 1)) such that f (y) = yn ηn for all y = hyn i ∈ lp . So, for any positive integer N , we have ∞ X |f (xn ) − f (x)| = (ξm,n − ξm )ηm m=1 !1/q !1/p N N X X |ηm |q |ξm,n − ξm |p ≤ m=1 m=1 ∞ X + !1/p |ξm,n − ξm |p m=N +1 ≤ N X !1/q ∞ X |ηm |q m=N +1 !1/p |ξm,n − ξn |p kηkq + 2M m=1 ∞ X !1/q |ηm |q . m=N +1 P∞ q 1/q < ε/(4M + 1), and then J so Given ε > 0, we can choose N so that m=N +1 |ηm | that |ξm,n − ξm | ≤ ε/(2N 1/p kηkq ) for n ≥ J and m ≤ N . Hence, for n ≥ J, we have |f (xn ) − f (x)| < ε. Since f and ε are arbitrary, it follows that hxn i converges to x. (c) Let f be a bounded linear functional on Lp . Then the Riesz Representation Theorem gives a function g ∈ Lq (with q = p/(p − 1)) such that Z f (x) = xg 1 2 for all x ∈ Lp . The solution of Problem 6.17 given on the web last semester shows that Z Z xg = lim xn g for any g ∈ Lq (because it only needed that xn → x in measure) and hence xn → x weakly. (d) As in Problem 6.17, let Z f (x) = x R (so f (x) = xg for g ≡ 1) and set xn = nχ(0,1/n) . Then kxn k1 = 1 and xn to0 in measure, but f (xn ) = 1 for all n while f (x) = 0. (e) If n 6= m, then kxn − xm k = 21/p , so hxn i is not a Cauchy sequence and hence it doesn’t converge in the strong topology. On the other hand, kxn kp = 1 for all n and ξm,n → 0 as n → ∞ for each fixed m, so x → 0 weakly by part (b). (f) To show that the distance between any two points in F is at least 1, we consider two cases. Let the two points be yn,m and yν,µ . First, if n = ν, then yn,m − yν,µ = n(xm − xµ ), so the distance between these two points is (2n)1/p ≥ 1. Second, if n 6= ν, then without loss of generality, we may assume that n < ν. Then the n-th term of yn,m − yν,µ must be 1, so the distance between the points is at least 1. Hence any nonconstant sequence in F isn’t Cauchy and therefore can’t converge. It follows that F is strongly closed. (g) To show that 0 is a weak closure point of F , we need to show that, for any ε > 0 and any finite number of functions f1 , . . . , fN in (lp )∗ , there is a choice of n and m such that |fi (yn,m )| < ε for all i. To proceed, we write zi = hζm,i i for the sequence in lq associated with fi . The n-th term test for convergence of sequences shows that there is an integer n such that |ζn,i | < ε/2 for all i and (from this n) a integer m such that |ζm,i | < ε/(2n) for all i. It follows that |fi (yn,m )| < ε for this choice of n and m. To show that there is no sequence hzj i from F that converges weakly to zero, we write zj = hζj,k i with ζj,k = yn(j,k),m(j,k) . If {n(j, k)} is bounded, then there is a an increasing sequence of positive integers hj(i)i such that n(j(i), k) is constant, say equal to ν. It follows that ζj(i),ν = 1 for all i and hence ζj,ν can’t converge to zero, so the sequence doesn’t converge weakly to zero by part (b). (h) No one actually gave a good solution, so I didn’t grade this one. The answer is rather tricky. If hxn i is a sequence that converges weakly in l1 to zero, then it converges strongly! To prove this, suppose to the contrary that hkxn k1 i does not converge to zero. Then there is a subsequence hxnj i and a positive number ε so that kxnj k1 > ε for all j. Let us set yj = xnj and write yj = hηm,j i. For any sequence z = hζm i in l∞ , we have ∞ ∞ X 1 X ηm,nj ζm ≤ ξm,nj ζm , ε m=1 m=1 and hence hyj i converges to with kyj k1 = 1 for all j. Since ky1 k = 1, there is a positive Pzero m1 integer m1 > 1 such that m=1 |ηm,1 | > 3/4. From the proof of part (b), for each m, the sequence P 1of numbers hηm,j i converges to zero. Hence there is a positive integer n1 > 1 such that m m=1 |ηm,n1 | < 1/4. Then, there is a positive integer m2 > m1 such that m2 X m=m1 +1 3 |ηm,n1 | > , 4 3 and a positive integer n2 > n1 such that m2 X m=1 1 |ηm,n2 | < . 4 We now proceed by induction to obtain two increasing sequences of positive integers hmj i and hnj i so that mk+1 X 3 |ηm,nk | > , 4 m=mk +1 and mk+1 X m=1 1 |ηm,nk+1 | < . 4 Now we consider the sequence z = hζm i ∈ l∞ with ζm = sgn xm,nk−1 for m in the interval (mk−1 , mk . Then ∞ X mk+1 X ηm,nk ζm = m=1 |ηm,nk | + mk X m=mk +1 m=1 mk+1 mk X X ≥ |ηm,nk | − m=mk +1 X m=mk +1 ∞ X ηm,nk ζm m=mk+1 +1 |ηm,nk | − m=1 mk+1 =2 ηm,nk ζm + ∞ X |ηm,nk | m=mk+1 +1 3 1 |ηm,nk | − kynk k1 > 2 −1= . 4 2 Hence lim k→∞ and therefore lim n→∞ ∞ X ηm,nk ζm 6= 0 m=1 ∞ X ηm,n ζm 6= 0, m=1 which contradicts the weak convergence of hxn i to zero. Now, we turn to weak∗ convergence. Since the functionals fn from part (b) come from sequences in c0 , it follows that, if a sequence hxn i in l1 converges weak∗ to 0, then hkxn k1 i is bounded and ξm,n → 0 for each m. Conversely, suppose hxn i is a sequence in l1 such that hkxn k1 i is bounded and ξm,n → 0 for each m. We want to show that ∞ X ξm,n ηm → 0 m=1 as n → ∞ for any y = hηm i in c0 , so let y ∈ c0 and ε > 0 be given. Also, suppose that M is a positive number such that kxn k1 ≤ M for all n. It follows that |ξn,m | ≤ M for all n and m. Then there is a positive integer N such that |ηm | ≤ ε/(2M ) for all m ≥ N . It follows that ∞ N X X ε | ξm,n ηm | ≤ |ξm,n ηm | + 2 m=1 m=1 4 P for all n. Now set M1 = N m=1 |ξm | and choose N1 so large that |ξm,n | < ε/(2M1 + 1) for n ≤ N and m ≥ N1 . It follows that ∞ X | ξm,n ηm | ≤ ε m=1 for m ≥ N1 , which implies that hxn i converges weak∗ to zero. 43. Let x and y be points in K and let N be an open neighborhood of 0. By Proposition 16, there is an open neighborhood U of 0 such that U + U ⊂ N and αU ⊂ U if 0 < α ≤ 1. Then there are points x0 and y 0 in K such that x − x0 and y − y 0 are elements of U . It follows that (λx + (1 − λ)y) − (λx0 + (1 − λ)y 0 ) = λ(x − x0 ) + (1 − λ)(y − y 0 ) ∈ λU + (1 − λ)U ⊂ U + U ⊂ N for any λ ∈ (0, 1). It follows that λx + (1 − λ)y ∈ K and hence K is convex.