78 CHAPTER 3. BASES IN BANACH SPACES Remark. Note that if aj = 1, for j 2 N, then in c0 sup n2N n X aj ej j=1 c0 = 1, P but the series j2N aj ej does not converge in c0 . Considering X as a subspace of X ⇤⇤ (via the canonical embedding) the image of X under T is the space of sequences 1 n o X Z := (ai ) 2 Y : aj ej converges in X . j=1 Proof of Proposition 3.3.6. Let K denote the basis constant of (en ), (e⇤n ) the coordinate functionals, and (Pn ) the canonical projections. It is straightforward to check that Y is a vector space and that ||| · ||| is a norm on Y . For x⇤ 2 X ⇤ and x⇤⇤ 2 X ⇤⇤ we have by Proposition 3.3.1 Pn⇤ (x⇤ ) = n X j=1 hx⇤ , ej ie⇤j and D hPn⇤⇤ (x⇤⇤ ), x⇤ i = x⇤⇤ , n X j=1 n E DX E hx⇤ , ej ie⇤j = hx⇤⇤ , e⇤j iej , x⇤ , j=1 which implies that (3.6) ⇤⇤ |||T (x )||| = sup n2N n X j=1 hx⇤⇤ , e⇤j iej = sup kPn⇤⇤ (x⇤⇤ )k Kkx⇤⇤ k. n2N Thus T is bounded and kT k K. Assume that (an ) 2 Y . We want to find x⇤⇤ 2 X ⇤⇤ , so that T (x⇤⇤ ) = (an ). Put n X ⇤⇤ xn = aj ej , for n 2 N. j=1 (where we identify X with its canonical image in X ⇤⇤ and, thus, ej with (ej ) 2 X ⇤⇤ ) Since kx⇤⇤ n kX ⇤⇤ = n X j=1 aj ej X |||(ai )|||, for all n 2 N, 3.3. SHRINKING, BOUNDEDLY COMPLETE BASES 79 and since X ⇤ is separable (and thus (BX ⇤⇤ , (X ⇤⇤ , X ⇤ )) is metrizable by ⇤ ⇤⇤ Exercise 8 in Chapter 2 ) (x⇤⇤ n has a w -converging subsequence xnj to an ⇤⇤ element x with kx⇤⇤ k lim sup kx⇤⇤ n k |||(aj )|||. n!1 It follows for m 2 N that ⇤ hx⇤⇤ , e⇤m i = lim hx⇤⇤ n j , em i = a m , j!1 and thus it follows that T (x⇤⇤ ) = (aj ), and thus that T is surjective. Finally, since (e⇤n ) is a basis for X ⇤ it follows for any x⇤⇤ n X |||T (x⇤⇤ )||| = sup n2N = = j=1 hx⇤⇤ , e⇤j iej sup n2N,x⇤ 2BX ⇤ ⌦ n X j=1 ⇤⇤ hx⇤⇤ , e⇤j ihx⇤ , ej i sup sup x , Pn⇤ (x⇤ ) x⇤ 2BX ⇤ n2N ⇤⇤ ↵ kx k (since Pn⇤ (x⇤ ) ! x⇤ if n ! 1), which proves that T is an isomorphism, and, that |||T (x⇤ )||| kx⇤⇤ k, for x⇤⇤ 2 X ⇤⇤ . Together with (3.6) that shows T is an isometry if K = 1. Lemma 3.3.7. Let X be a Banach space with a basis (en ), with basis constant K and let (e⇤n ) be its coordinate functionals. Let Z = span(e⇤n : n 2 N) ⇢ X ⇤ and define the operator S : X ! Z ⇤, x 7! (x)|Z i.e. S(x)(z) = hz, xi, for z 2 Z and x 2 X. Then S is an isomorphic embedding of X into Z ⇤ and for all x 2 X. 1 kxk kS(x)k kxk. K Moreover, the sequence (S(en )) ⇢ Z ⇤ are the coordinate functionals of (e⇤n ) (which by Proposition 3.3.1 is a basis of Z). Proof. For x 2 X note that kS(x)k = sup z2Z,kzkX ⇤ 1 |hz, xi| sup |hx⇤ , xi| = kxk, x⇤2BX ⇤ By Corollary 1.4.6 of the Hahn Banach Theorem. 80 CHAPTER 3. BASES IN BANACH SPACES On the other hand, again by using that Corollary of the Hahn Banach Theorem, we deduce that kxk = sup |hw⇤ , xi| ⇤ w⇤ 2BX = ⇤lim ⇤ lim |hw⇤ , Pn (x)i| w 2BX n!1 = ⇤lim ⇤ lim |hPn⇤ (w⇤ ), xi| w 2BX n!1 sup sup |hPn⇤ (w⇤ ), xi ⇤ n2N w⇤ 2BX sup sup n2N z2span(e⇤j :jn),kzkK |hz, xi = KkS(x)k. Theorem 3.3.8. Let X be a Banach space with a basis (en ), and let (e⇤n ) be its coordinate functionals. Let Z = span(e⇤n : n 2 N) ⇢ X ⇤ . Then the following are equivalent a) X is isomorphic to Z ⇤ , via the map S as defined in Lemma 3.3.7 b) (e⇤n ) is a shrinking basis of Z. c) If (aj ) ⇢ K, with the property that sup n2N then P1 j=1 aj ej n X j=1 aj ej < 1, converges. In that case we call (en ) boundedly complete. Proof. “(a))(b)” Assuming condition (a) we will verify condition (b) of Theorem 3.3.4 for Z and its basis (e⇤n ). So let z ⇤ 2 Z ⇤ . By (a) we can write z ⇤ = S(x) for some x 2 X. Since x = limn!1 Pn (x), where (Pn ) are the canonical projection for (en ), we deduce that sup w2span(e⇤j :j>n),kwk1 hz ⇤ , wi = = = sup hS(x), wi sup hw, xi sup hw, (I w2span(e⇤j :j>n),kwk1 w2span(e⇤j :j>n),kwk1 w2span(e⇤j :j>n),kwk1 Pn )(x)i 3.3. SHRINKING, BOUNDEDLY COMPLETE BASES k(I 81 Pn )(x)k !n!1 0. It follows now from Theorem 3.3.4 that (e⇤j ) is a shrinking basis of Z. “(b))(c)” Assume (b) and let (aj ) ⇢ K so that n X |||(aj )||| = sup n2N aj ej = sup n2N j=1 n X j=1 aj (ej ) < 1. As in Proposition 3.3.6 we can show that there is an x⇤⇤ 2 X ⇤⇤ , so that hx⇤⇤ , e⇤n i = an for all n 2 N (for that part, the assumption that (en ) was shrinking was not used). Let z ⇤ be the restriction of x⇤⇤ to the space Z. Since (e⇤j ) is a shrinking basis of Z, since by Lemma 3.3.7 (S(en )) are the coordinate functionals we can write (in a unique way) z⇤ = 1 X bj S(ej ) j=1 hx⇤⇤ , e⇤j i But this means that aj = = hz ⇤ , e⇤j i = bj for all j 2 N, P and since 1 S is an isomorphism between X and its image it follows that j=1 aj ej converges in norm in X. “(c)) (a)” By Lemma 3.3.7 it is left to show that the operator S is surjective. Thus, let z ⇤ 2 Z ⇤ . Since (e⇤n ) is a basis of Z and (S(en )) ⇢ Z ⇤ are the coordinate functionals of (e⇤n ), it follows from Proposition 3.3.1 that z ⇤ is the w⇤ limit of (zn⇤ ) where zn⇤ = n X j=1 Since w⇤ -converging hz ⇤ , e⇤j iS(ej ). sequences are bounded it follows that n X sup n2N j=1 and, thus, by Lemma 3.3.7 sup n2N hz ⇤ , e⇤j iS(ej ) < 1 n X j=1 hz ⇤ , e⇤j iej < 1. By our assumption (c) it follows therefore that x = in X, and moreover S(x) = lim n!1 which proves our claim. n X j=1 Pn j=1 hz hz ⇤ , e⇤j iS(ej ) = z ⇤ , ⇤ , e⇤ ie j j converges 82 CHAPTER 3. BASES IN BANACH SPACES Theorem 3.3.9. Let X be a Banach space with a basis (en ). Then X is reflexive if and only if (ej ) is shrinking and boundedly complete, or equivalently if (ej ) and (e⇤j ) are shrinking. Proof. Let (e⇤n ) be the coordinate functionals of (en ) and (Pn ) be the canonical projections for (en ). “)” Assume that X is reflexive. By Proposition 3.3.1 it follows for every x⇤ 2 X ⇤ n X ⇤ ⇤ x =w lim Pn⇤ (x⇤ ) = w lim Pn⇤ (x⇤ ), n!1 n!1 j=1 w which implies that x⇤ 2 span(e⇤n : n 2 N) , and thus, by Proposition 2.2.5 k·k k·k x⇤ 2 span(e⇤n : n 2 N) . It follows therefore that x⇤ = span(e⇤n : n 2 N) and thus that (ej ) is shrinking (by Proposition 3.3.1). Thus X ⇤ is a Banach space with a basis (e⇤j ) which is also reflexive. We can therefore apply to X ⇤ what we just proved for X and deduce that (e⇤n ) is a shrinking basis for X ⇤ . But, by Theorem 3.3.8 (in this case Z = X ⇤ ) this means that (en ) is boundedly complete. “(” Assume that (en ) is shrinking and boundedly complete, and let x⇤⇤ 2 X ⇤⇤ . Then X ⇤⇤ = (X ⇤⇤ , X ⇤ ) " n!1 n X j=1 hx⇤⇤ , e⇤j i (ej ) By Proposition 3.3.1 and the fact that X ⇤ = span(e⇤j : j 2 N) has (e⇤j ) as a basis, since (ej ) is shrinking =k·k lim lim n!1 n X j=1 # hx⇤⇤ , e⇤j i (ej ) 2 (X) P Since supn2N k nj=1 hP ⇤⇤ (x⇤⇤ ), e⇤j iej k < 1, by Exercise 9 in Chapter 2,and since (ej ) is boundedly complete which proves our claim. The last Theorem in this section describes by how much one can perturb a basis of a Banach space X and still have a basis of X. Theorem 3.3.10. (The small Perturbation Lemma) Let (xn ) be a basic sequence in a Banach space X, and let (x⇤n ) be the ⇤ coordinate functionals (they are elements of span(xj : j 2 N) ) and assume 3.3. SHRINKING, BOUNDEDLY COMPLETE BASES 83 that (yn ) is a sequence in X such that (3.7) c= 1 X n=1 yn k · kx⇤n k < 1. kxn Then a) (yn ) is also basic in X and isomorphically equivalent to (xn ), more precisely (1 c) 1 X n=1 1 X an xn n=1 for all in X converging series x = an yn (1 + c) P 1 X an xn , n=1 n2N an xn . b) If span(xj : j 2 N) is complemented in X, then so is span(yj : j 2 N). c) If (xn ) is a Schauder basis of all of X, then (yn ) is also a Schauder basis of X and it follows for the coordinate functionals (yn⇤ ) of (yn ), that yn⇤ 2 span(x⇤j : j 2 N), for n 2 N. Proof. By Corollary 1.4.4 of the Hahn Banach Theorem we extend the functionals x⇤n to functionals x̃⇤n 2 X ⇤ , with kx̃⇤n k = kx⇤n k, for all n 2 N. Consider the operator: T : X ! X, P1 x 7! yn k · kx⇤n k 1 X n=1 hx̃⇤n , xi(xn yn ). Since n=1 kxn < 1, T is well defined, linear and bounded and kT k c < 1. It follows S = Id T is an isomorphism between X and it self. Indeed, for x 2 X kS(x)k kxk kT k · kxk (1 c)kxk and P we have, n (y) (T 0 = Id) then if y 2 X, define x = 1 T n=0 (Id T )(x) = 1 X n=0 T n (y) T 1 ⇣X n=0 1 ⌘ X T n (y) = T n (y) n=0 1 X T n (y) = y. n=1 Thus Id T is surjective, and, it follows form Corollary 1.3.6 that Id T is an isomorphism between X and itself. (a) We have (I T )(xn ) = yn , for n 2 N, this means in particular that (yn ) is basic and (xn ) and (yn ) are isomorphically equivalent. (b) Let P : X ! span(xn : n 2 N) be a bounded linear projection onto span(xn : n 2 N). Then it is easily checked that Q : X ! span(yn : n 2 N), x 7! (Id T) P (Id T) 1 (x), 84 CHAPTER 3. BASES IN BANACH SPACES is a linear projection onto span(yn : n 2 N). (c) If X = span(xn : n 2 N), then, since I T is an isomorphism, (yn ) = ((I T )(xn )) is also a Schauder basis of X. Moreover define for k and i in N, ⇤ y(i,k) = k X j=1 hyi⇤ , xj ix⇤j = n X j=1 h (xj ), yi⇤ ix⇤j 2 span(x⇤j : j 2 N). ⇤ It follows from Proposition 3.3.1, part (b), that w⇤ limk!1 y(i,k) = yi⇤ , P ⇤ ⇤ which implies that yi⇤ (x) = 1 j=1 hyi , xj ihxj , xi, for all x 2 X, and thus for k i kyi⇤ ⇤ y(i,k) k = sup |hyi⇤ ⇤ y(i,k) , xi| x2BX = sup x2BX = sup x2BX kyi⇤ k 1 X j=k+1 1 X j=k+1 1 X j=k+1 hyi⇤ , xj ihx⇤j , xi hyi⇤ , xj kxj yj ihx⇤j , xi yj k · kx⇤j k ! 0, if k ! 1. ⇤ so it follows that yi⇤ = k · k limk!1 y(k,i) 2 span(x⇤j : j 2 N) for every i 2 N, which finishes the proof of our claim (c). Exercises 1. Prove that Y with ||| · |||, as defined in Proposition 3.3.6 is a normed linear space. 2. A Banach space X is said to have the Approximation Property if for every compact set K ⇢ X and every " > 0 there is a finite rank operator T so that kx T (x)k < " for all x 2 K. Show that every Banach space with a Schauder basis has the approximation property. 3.3. SHRINKING, BOUNDEDLY COMPLETE BASES 3. 85 Show that (ei ) is a shrinking basis of a Banach space X, then the coordinate functionals (e⇤i ) are boundedly complete basis of X ⇤ . 4.⇤ A Banach space is called L(p, ) -space, for some 1 p 1 and some 1, if for every finite dimensional subspace F of X and every " > 0 there is a finite dimensional subspace E of X which contains F and so dim(E) that dBM (E, `p ) < + ". Show that Lp [0, 1], 1 p < 1 is a L(p,1) -space. Hint: Firstly, he span of the first n elements of the Haar basis is isometrically isomorphically to `np (why?), secondly consider Small Perturbation Lemma.