Math 516 Professor Lieberman January 31, 2005 HOMEWORK #2 SOLUTIONS Chapter 10 20. First, we let f be the linear functional defined on c (the space of all sequences with limits) by f (hxn i) = lim xn , define p on l∞ by p(hxn i) = lim sup xn , and let G be the semigroup of linear operators of the form A such that there is a positive integer M such that Ahxn i = hxn+M i. (Of course, the integer M will be different for different elements of G.) Note that f (Ax) = f (x) and p(Ax) = p(x) and that f (x) = p(x) for all x ∈ c. Proposition 10.5 then implies that there is a linear functional F on l∞ which extends f . The second inequality in Property i then follows because F ≤ p while the first inequality follows because F [hξn i] = −F [h−ξn i] ≥ − lim sup(−ξn ) = lim inf ξn . Properties ii and iii are just the linearity of F and Property iv is the statement F (Ax) = F (x) for any A ∈ G and any x ∈ l∞ . 21. Let X be the vector space of all bounded functions defined on R which are zero outside of some finite interval, and let S be the subspace of X consisting of all measurable functions which are elements of X. We then define the linear functional f on S by Z f (x) = x(t) dt, and we define p on X by setting Z p(x) = inf{ y : y ∈ L∞ , x ≤ y}. R (For brevity, we write x for this infimum, in analogy to the upper Riemann integral.) Finally, we write G for the semigroup of all translation operators, so A ∈ G if there is a number s such that Ax(t) = x(t + s). It’s easy to check that the hypotheses of the HahnBanach theorem are satisfied in this case. In particular,Rthe subadditivity of p follows by letting ε > 0 and taking y1 and y2 so that xi ≤ yi and yi ≤ p(xi ) + ε/2 for i = 1, 2. It follows that Z p(x1 + x2 ) ≤ y1 + y2 ≤ p(x1 ) + p(x2 ) + ε and hence p(x1 + x2 ) ≤ p(x1 ) + p(x2 ). Hence, there is a linear functional F defined on X which is an extension of f . Let us also note that Z Z F (x) = −F (−x) ≥ − − x(t) dt = x(t) dt. Hence, if x is nonnegative, then so is F (x). We now define µ by µ(A) = F (χA ). We now check the claimed properties for µ. First, if A ∩ B = ∅, then χA∪B = χA + χB , so µ(A ∪ B) = F (χA ) + F (χB ) = µ(A) + µ(B). 1 2 Next, property ii. is a consequence of the invariance of F under the semigroup G. Furthermore, if A ⊂ B, then we can write B = A ∪ C with C = B ∼ A, and we see that A ∩ C = ∅. Therefore, µB = µA + µC ≥ µA because χC is nonnegative and we Rjust showed that µC = F (χC ) ≥ 0. Finally, if A is Lebesgue measurable, then µ(A) = χA , which is just the Lebesgue measure of A. 22. First, write ϕ for the natural isomorphism of X into X ∗∗ and ϕ∗ for the natural isomorphism of X ∗ into X ∗∗∗ . If X is reflexive, let y ∈ X ∗∗∗ and define ψ : X ∗∗∗ → X ∗ by (ψy)(x) = y(ϕx). But, for any x ∈ X, we have (ϕ∗ (ψy))(ϕx) = (ϕx)(ψy) = (ψy)(x), so y = ϕ∗ (ψy). Hence the image of ϕ∗ is all of X ∗∗∗ and therefore X ∗ is reflexive. If X is not reflexive, then there is x∗∗ ∈ X ∗∗ ∼ ϕ(X). Since X is a Banach space, ϕ(X) is closed, and hence there is a positive number δ such that d(x∗∗ , ϕ(X)) ≥ δ. Then Proposition 6 shows that there is z ∈ X ∗∗ such that z(ϕx) = 0 for all x ∈ X but z(x∗∗ ) = δ. If y ∈ X ∗ , then (ϕ∗ y)(ϕx) = (ϕx)(y) = y(x), so, if (ϕ∗ y)(ϕx) = 0 for all x ∈ X, it follows that y = 0. Hence z cannot be ϕ∗ y for any y ∈ X ∗ , which means that X ∗ is also not reflexive. 26. Take T to be the family of all operators Tn . For each x, hTn xi is a convergent sequence, so it’s bounded. Proposition 11 then implies that there is a constant M such that kTn xk ≤ M kxk for all x ∈ X and all n. Sending n → ∞, we see that kT xk ≤ M kxk and hence T is bounded. 29. (a) (Ignore the hint). Let X be the vector space C 1 [0, 1] with norm kxk = sup x, let Y 0 be the vector space C[0, 1] with norm kyk = sup y, and pdefine Ax = x . Then X is not a Banach space: The sequence hxn i defined by x+n(t) = x + 1/n converges uniformly to x, defined by x(t), but xn ∈ X and x ∈ / X. This sequence also shows that A is unbounded. But if hxn i is a convergent sequence in X such that hAxn i converges in Y , then hx0n i converges uniformly to some continuous function y and hxn i converges uniformly to some continuous function x. It follows from advanced calculus that x0 = y and hence A is closed. (b) (From Real Variables by Alberto Torchinsky, but that isn’t the original source of this example) Let (V, k · k) be an infinite-dimensional Banach space. Then, as shown in 515, there is a vector space basis B for V . (It can be obtained via the Hausdorff maximal principle as a maximal linearly independent subset of V .) Define H = {h : h = b , b ∈ B}, kbk and define a new norm k · k1 on V by kxk1 = n X |ai |, i=1 where x= n X ai hi , hi ∈ H. i=1 (In other words, the ai ’s are the coefficients for x with respect to this basis.) Then we write X for the Banach space (V, k · k) and Y for the Banach space (V, k · k1 ), and we define 3 A : X → Y by Ax = x. To see that A is closed, let ixn h be a sequence in V and suppose that xn → x in X and Axn → y in Y . First, we observe that n n X X kxk = ai hi ≤ |ai |khi k = kxk1 . i=1 i=1 Hence kxn − yk = kAxn − yk ≤ kAxn − yk1 , so xn → y in X and therefore x = y. To show that A is unbounded, we only have to show that Y is not complete by virtue of Proposition 10. Let hhi i be a sequence of distinct vectors in H and let n X xn = 2−i hi . i=1 Then it’s easy to check that hxn i is a Cauchy sequence in Y . Suppose now that this sequence had a limit x. Then x must be a linear combination of the hi ’s in our sequence (because the component of each xn with respect to any other h is zero), but the component with respect to hi must be nonzero (in fact 2−i ) for each i, so x is not a linear combination of the hi ’s. It follows that Y is not a Banach space, so A is unbounded.