Chapter 4 Fundamental Theorems For Normed And Banach Spaces 4.1 Zorn’s Lemma 4.1-1 Definition. A partially ordered set is a set M with a partial order relation satisfying the following: (PO1) a a for all aM. (PO2) If a b and b a, then a = b. (PO3) If a b and b c, then a c. 4.1-2 Definition. An upper bound of a subset W of a partially ordered set M is uM such that x u for all xM. A maximal element of M is mM such that if m x, then m = x. A subset C of M is called a chain or a totally ordered set if for any a, bC, a b or b a ( or both ). 4.1-3 Zorn’s Lemma. Let M be a non empty partially ordered set. If every chain C M has an upper bound, then M has at least one maximal element. 4.1-4 Theorem. ( Total Orthonormal Set ) In every Hilbert space H {0} there exists a total orthonormal set. Proof. Let M be the set of all orthonormal subsets of H. Since H {0}, then there is xH and x 0. Hence { x || x|| } is an orthonormal subset of H. Hence M . Also note that (M, ) defines a partially ordered set ( set inclusion). Let C M be any chain, then X = A is an upper bound of C. By Zorn’s Lemma AC M has a maximal element F. Suppose that F is not total in H. Then by Theorem z 3.6-2, F {0} and so there is zF and z 0. Then F1 = F { || z || } is orthonormal and F is a proper subset of F1 which is a contradiction of maximality of F. Therefore, the orthonormal se F is total in H 4.2 Hahn-Banach Theorem 4.2-1 Definition. A sublinear functional is a real valued function p on a vector space X which is a) subadditive, p(x + y) p(x) + p(y) for all x, yX. b) positive homogeneous, p(x ) = p(x) for all xX and all 0. 4.2-2 Hahn-Banach Theorem. ( Extension of Linear Functionals ) Let X be a real vector space and p is a sublinear functional on X. Furthermore, let f be a linear functional which is defined on a subspace Z of X and satisfies ~ f(x) p(x) for all xZ. Then f has a linear extension f from Z to X satisfying ~ f (x) p(x) for all xX. Proof. (a) Let E be the set of all linear extensions g of f that satisfy g(x) p(x) for all xD(g). Since fE, then E . Define on E a partial ordering by g h if and only if h is an extension of g. For any chain C E, define g by g (x) = g(x) if xD(g) and gC. Then g is a linear functional and D( g ) = 1 gC D( g ) which is a vector space since C is a chain (why?). It is clear that g g for all gC. Then g is an upper bound of C. Hence by Zorn’s Lemma E has a ~ maximal element say, f . By the definition of E this is a linear extension of f ~ ~ which satisfies f (x) p(x) for all xD( f ). ~ ~ ~ b) We show that D( f ) = X. Suppose that D( f ) X. Choose y1X\D( f ) and ~ ~ let Y1 be the subspace spanned by D( f ) and y1. Since 0D( f ), then y1 0. ~ Any xY1 can be written uniquely as x = y + y1, yD( f ). To see this ~ ~ suppose that x = y + y1 and x = y + y1. Then ( - ) y1 = y - y. However, ~ ~ ~ ~ y1D( f ) and y - yD( f ), then - = 0, so that = and y = y. This ~ proves the uniqueness. Define g1 : Y1 R by g1(y + y1) = f (y) + c, where c is any real constant. It is easy to see that g1 is linear ( How?). If = 0 then ~ ~ ~ g1(y) = f (y) for all yD( f ). Hence g1 is a proper extension of f . However, g1(x) p(x) for all x = y + y1D(g1), [see its proof below] ..………..……(1) ~ ~ Hence g1E which is contradicts the maximality of f . Therefore, D( f ) = X. ~ ~ ~ (c) Finally we prove (1). Let y, z D( f ). For any fixed y1X, f (y) - f (z) = ~ f (y - z) p(y - z) = p(y + y1 - y1- z) p(y + y1) + p(- y1- z). Then ~ ~ ~ ~ -p(-y1- z) - f (z) p(y + y1) - f (y). Hence u = sup{-p(-y1- z)- f (z): zD( f )} ~ ~ inf{ p(y + y1) - f (y): yD( f )} = w. Then there is a real number c such that ~ ~ u c w. So that -p(-y1- z) - f (z) c for all zD( f ) .………………..(2), ~ ~ and c p(y + y1) - f (y) for all yD( f ) ..……………..………………..(3) To prove (1) we have three steps. First step, when < 0. Put z = -1y in (2), ~ then multiply both sides by - to get, p(-y1- -1y) + f (y) -c. Then, for ~ x = y + y1 g1(x) = f (y) + c -p(-y1- -1y) = p(y + y1) = p(x) since p is sublinear. Thus (1) is true when < 0. Second step, when = 0. Then x = ~ ~ yD( f ) and g1(x) = f (y) p(y) = p(x). Thus (1) is true when = 0. Third step, when > 0. By (3), but replace y by -1y, and then multiply both sides ~ ~ by to get c p( -1y + y1) - f ( -1y) = p(x) - f (y). Hence g1(x) = ~ f (y) + c p(x). Thus (1) is true when > 0. Therefore, (1) is true for all , that is g1(x) p(x) for all x = y + y1D(g1). This completes the proof H. W. 5-10. H.W.* 8, 10. 2 4.3 Hahn-Banach TheoremFor Complex Vector Spaces And Normed Spaces 4.3-1 Hahn-Banach Theorem. ( Generalized ) Let X be a real or a complex vector space and p is a subadditive real-valued functional on X, and for any scalar , p(x) = ||p(x) ……………………………………………………...(1) Furthermore, let f be a linear functional which is defined on a subspace Z of X and satisfies |f(x)| p(x) for all xZ…………………………………………(2) ~ ~ Then f has a linear extension f from Z to X satisfying | f (x)|p(x) for all xX. Proof. Left to the reader. 4.3-2 Hahn-Banach Theorem. ( Normed Space ) Let f be a bounded linear functional on a subspace Z of a normed space X. Then there exists a bounded ~ linear functional f on X which is an extension of f to X and has the same ~ ~ ~ norm, || f ||X = || f || Z, where || f ||X = sup { | f (x)| : xX, || x || = 1 }, || f || Z = sup{ | f(x) | : xZ, || x || = 1}, and || f || Z = 0 in the trivial case Z = {0}. ~ Proof. If Z = {0}, then f = 0 and the extension is f = 0. Let Z {0}. Then for any xZ we have | f(x) | || f ||Z || x ||. Define p : X R by p(x) = || f ||Z || x ||. Then | f(x) | p(x) for all xZ. Furthermore, for any x, yX and any R we have, (1) p( x + y ) = || f ||Z || x + y || || f ||Z ( || x || + || y || ) = p( x ) + p( y ) and (2) p(x ) = || f ||Z || x || = | | p( x ). Hence by Theorem 4.3-1, there ~ exists a linear functional f on X which is an extension of f and satisfies ~ ~ | f (x)| p(x) = || f ||Z || x || for all xX. Then || f ||X || f ||Z. However, || f ||Z ~ ~ ~ = || f ||Z || f ||X. Hence || f ||X = || f || Z 4.3-3 Theorem. ( Bounded Linear Functional ) Let X be a normed space and let x0 0 be any element of X. Then there exists a bounded linear functional ~ ~ ~ f on X such that || f || = 1 and f ( x0 ) = || x0 ||. Proof. Consider the subspace Z = { x0 : is a scalar }. Define f : Z k by f(x) = f(x0) = || x0 ||, where k is the scalar field ( real or complex ). It is easy to see that f is linear (how?). Also f is bounded, where |f(x)| = |f(x0)| = ||x0|| = || x || for all xX. Hence || f || = 1. By Theorem 4.3-2, f has a linear extension ~ ~ ~ f from Z to X of norm || f || = || f || = 1. Moreover, f ( x0 ) = f( x0 ) = || x0 || 4.3-4 Corolllary. For any x in a normed space X we have, || x || = sup{ | f ( x )| || f || : fX / , f 0 }. Hence if x0 is such that f( x0 ) = 0 for all fX /, then x0 = 0. ~ ~ ~ Proof. By Theorem 4.3-3, there exists f X / such that || f || = 1 and f ( x ) = || x || where x is a nonzero element in X. So that sup{ 3 | f ( x )| || f || : fX / , f 0 } ≥ ~ | f ( x )| ~ || f || = || x ||. However, | f(x) | ≤ || f || || x ||, then sup{ || x ||. Therefore, sup{ | f ( x )| || f || | f ( x )| || f || : fX / , f 0 } ≤ : fX / , f 0 } = || x ||. If x0 is such that f( x0 ) = 0 for all fX /, then || x0 || = sup{ | f ( x )| || f || : fX / , f 0 } = . Hence x0 = 0 H. W. 1, 2, 4, 5, 8, 11, 15. H.W.* 4. 4.5 Adjoint operator 4.5-1 Definition. Let T : X Y be a bounded linear operator, where X and Y are normed spaces. Then the adjoint operator T ×: Y / X / of T is defined by (T ×g)(x) = g(Tx) = f(x), where X / and Y / are the dual spaces of X and Y, respectively, and xX, fX / and gY /. 4.5-2 Theorem. The adjoint operator of T in the above definition is linear and bounded with || T × || = || T ||. Proof. For any xX, g1, g2Y / and any scalar we have, (T × (g1 + g2))(x) = (g1 + g2)(Tx) = g1(Tx) + g2(Tx) = (T ×g1)(x) + (T ×g2)(x). Hence T ×(g1 + g2) = T ×g1 + T ×g2. Therefore, T is linear. Since gY / and T are bounded then, | f(x) | = | g(Tx) | ≤ || g || || Tx || ≤ || g || || T || || x || and so || f || ≤ || g || || T ||. By the definition of T ×, T ×g = f, and so || T ×g || = || f || ≤ || g || || T ||. Hence || T × || = sup{|| T ×g || : gY /, || g || = 1} ≤ || T ||……………………(1) This means that T × is bounded. By Theorem 4.3-3 for any x0 0 in X there exists g0Y / such that || g0 || = 1 and g0( Tx0 ) = || Tx0 ||. Let f0 = T ×g0. Then || Tx0 || = g0( Tx0 ) = f0( x0 ) ≤ || f0 || || x0 || = || T ×g0 || || x0 || ≤ || T ×|| || g0 || || x0|| = || T ×|| || x0||. This implies that || T || ≤ || T × ||. By this and (1) we have, || T ×|| = || T || 4.5-3 Theorem. If S, T : X Y and W : Y Z are bounded linear operators, where X, Y and Z are normed spaces. Then (1) ( S + T ) × = S× + T × and (T)× = T× for any scalar . (2) ( WT ) × = T×W×. (3) If T -1BL(Y, X), then (T ×) -1BL(X/, Y/ ) and (T ×) -1 = (T -1) ×. Proof. Left to the reader. 4.5-4 Theorem. Let H1 and H2 be two Hilbert spaces, T BL(H1, H2). Then there exist A1 : H/1 H1 and A2 : H/2 H2 such that T* = A1T×A2-1 where T* and T× are the Hilbert adjoint and the adjoint operators of T, respectively, and both A1 and A2 are bijective, isometric and conjugate linear. Proof. Let T : H1 H2 be a bounded linear operator and T× : H/2 H/1 be its adjoint. Then T×g = f where g(Tx) = f(x), fH/1, gH/2 and xH1. Then by 4 Theorem 3.8-1 there exist a unique x0H1 and a unique y0H2 such that f(x) = < x, x0>, || f || = || x0 ||, g(y) = < y, y0> and || g || = || y0 ||. By noting that x0 and y0 are uniquely determined by f and g, respectively we can define A1: H/1 H1 and A2 : H/2 H2 by A1f = x0 and A2g = y0. Then || A1f || = || x0 || = || f ||. Hence A1 is isometric and so it is one to one. Let hH1. Then f : H1 k defined by f(x) = < x, h > for all xH1 is a bounded linear operator ( k is the scalar field). Then A1f = h, hence A1 is onto. Similarly, for A2. To show that A1 is conjugate linear, let f1, f2 H/1 and any scalar. Then there exist x1, x2H1 such that f1(x) = < x, x1 > and f2(x) = < x, x2 > for all xH1. Then (f1 + f2)(x) = f1(x) + __ __ f2(x) = < x, x1 > + < x, x2 > = < x, x1+ x2 >. Hence A1(f1 + f2) = x1+ x2 __ = A1f1 + A1f2. Therefore, A1 is conjugate linear. Similarly for A2. For any y0H2 there exists gH/2 such that A2g = y0, so A2-1y0 = g. Then (A1T×A2-1)(y0) = (A1T×)(g) = A1f = x0. Also, < Tx, y0 > = g(Tx) = f(x) = < x, x0 > = < x, (A1T×A2-1)(y0) >. However, < Tx, y0 > = < x, T*y0 > and T* is unique, hence T* = A1T×A2-1 H. W. 1-5, 8,9 H.W.* 8. 4.6 Reflexive Spaces 4.6-1 Lemma. For every fixed x in a normed space X, the functional gx defined on X/ by gx(f) = f(x) ( for all fX/ ) is a bounded linear functional so that gxX// and has the norm || gx || = || x ||. Proof. Left to the reader. 4.6-2 Lemma. Let X be a normed space. Then the canonical mapping C:XX// defined by C(x) = gx, where gx(f) = f(x) ( for all fX/ ) is an isomorphism from X onto R(C), the range of C. Proof. For any fX/, x, yX and any scalar , C(x + y)(f) = g x + y(f) = f(x + y) = f(x) + f(y) = gx(f) + gy(f) = C(x)(f) + C(y)(f). Hence C(x + y) = C(x) + C(y), and so C is linear. Note that for all fX/, g x – y (f) = f(x–y) = f(x) – f(y) = gx(f) – gy(f) = (gx–gy)(f). Hence g x – y = gx–gy. By Lemma 4.6-1, || C(x) – C(y) || = || gx– gy || = || g x – y || = || x – y ||. This means that C is isometric and so it is one to one. Therefore, C is an isomorphism from X onto R(C) 4.6-3 Definition. A normed space X is said to be reflexive if R(C) = X //, where C as in Lemma 4.6-2. Notes. (1) By Lemma 4.6-2, the normed space X is isomorphic to a subspace of X//. Hence X is embeddable in X//, and in this case C is called the canonical embedding of X into X//. (2) If the normed space X is reflexive, then it is isomorphic with X//. 5 4.6-4 Theorem. If a normed space X is reflexive, then it is complete. Proof. Since X is reflexive, then X is isomorphic to X//. However, X// is a Banach space, then X is a Banach space 4.6-5 Theorem. Every finite dimensional normed space is reflexive. Proof. Since dimX is finite, then X/ = X*. By Theorem 2.9-3 dimX* = dimX, then X// = X**. However, C : X X** is an isomorphism, so that C : X X// is an isomorphism. Hence X is reflexive Note. Since ( p ) // = p ( 1 < p < ), then p is reflexive. 4.6-6 Theorem. Every Hilbert space is reflexive. Proof. Left to the reader. 4.6-7 Lemma. Let Y be a proper closed subspace of a normed space X. Let ~ ~ x0X-Y be arbitrary and δ = inf { || y - x0 || : y Y}, the distance from x0 to Y. ~ ~ ~ Then there exists f X/ such that || f || = 1, f (y) = 0 for all yY and ~ f ( x0) = δ. Proof. Consider the subspace Z = { y + x0 : yY, is a scalar } of X, and define the functional f on Z by f(z) = f(y + x0) = δ. Then f is linear (how?) Since Y is closed, then δ > 0 and so f ≠ 0. Note that for all yY, f(y) = f(y + 0x0) = 0 and f(x0)=δ (let y = 0 and =1). Now we show that f is bounded. If = 0, then f(z) = 0. If ≠ 0, then - yY for all yY and | f(z)| = | f(y + x0) | 1 ~ ~ = ||δ = || inf{|| y - x0 || : y Y} || || - y - x0 || = || y + x0 || = || z ||. Hence f is bounded and || f || 1 ……………………………………………..(1) By the definition of infimum there is a sequence (yn) of elements in Y such that 1 |f ( y n x 0 )| || y n x 0 || || yn - x0 || δ as n . Then || f || = || y n x 0 || 1 as n . By this an (1) we have || f || = 1. By Hahn-Banach Theorem 4.3-2, there exists ~ ~ ~ ~ f X/ such that || f || = 1, f (y) = 0 for all yY and f ( x0) = δ 4.6-8 Theorem. If the dual space X/ of a normed space X is separable, then X itself is separable. Remark. A separable normed space X with a nonseparable dual space X/ can’t be reflexive. Proof. Suppose that X/ is nonseparable but X is separable and reflexive. Then X// isomorphic to X which implies the separability of X// . Then by Theorem 4.6-8, X/ is separable which is a contradiction. Therefore, X can’t be reflexive 1 Example. is not reflexive. 1 Proof. Since is separable but ( reflexive H. W. 1, 5, 7-9 H.W.* 8. 1 / ) = 6 is not separable, then 1 is not 4.7 Category Theorem. Uniform Boundedness Theorem 4.7-1 Definition. A subset M of a metric space X is said to be __ __ (a) rare ( or nowhere dense ) in X if M has no interior point ( ( M )0 = φ ), (b) meager ( or of first category ) in X if M is the union of countably many sets each of which is rare in X. (c) nonmeager ( or of the second category ) in X if M is not meager in X. 4.7-2 Bair Category Theorem. If a metric space X ≠ φ is complete then it is nonmeager in it self. Hence if X ≠ φ is complete and X = Ak , Ak is closed k 1 for all k then at least one Ak contains a nonempty open subset. Proof. Suppose that the complete metric space X ≠ φ is meager in itself. Then __ X= Mk with each Mk is rare ( nowhere dense ) in X. Then ( M 1 )0 = φ, that k 1 __ means M 1 does not contain a nonempty open set. But X does ( for example X __ __ __ itself ), so that M 1 ≠ X. Hence (~ M 1 ) = X - M 1 is nonempty and open, so __ we can choose p1(~ M 1 ) and an open ball about it, say B1 = B(p1; ε1 ) __ __ (~ M 1 ) and ε1 < ½ . Since M2 is nowhere dense in X, then M 2 does not __ contain a nonempty open set. Hence B(p1; ½ε1 ) is not a subset of M 2 which __ implies that (~ M 2 ) ∩ B(p1; ½ε1 ) is nonempty and open, so we can choose __ p2(~ M 2 ) ∩ B(p1; ½ε1 ) and an open ball about it, say B2 = B(p2; ε2 ) __ (~ M 2 ) ∩ B(p1; ½ε1 ) and ε2 < ½ ε1. So by induction we obtain a sequence of balls Bk = B(pk; εk ), εk < 2-k such that Bk+1 B(pk; ½εk ) Bk. Since εk < 2-k, then d ( pk , pk 1 ) ≤ k 1 2 k and so it converges. Hence (pk) is a Cauchy k 0 sequence in the complete metric space X so it converges, say lim pk = pX. Also for all m and all n > m we have, Bn B(pm; ½εm ) so that d(pm, p) ≤ d(pm, pn ) + d(pn, p) < ½ εm + d(pn, p) → ½ εm as n → ∞ Hence pBm for all m. __ Since Bm (~ M m ), then p Mm for all m, so that p Mk = X. This k 1 contradicts pX. Therefore, X must be nonmeager in it self 4.7-3 Uniform Boundedness Theorem. Let (Tn ) be a sequence of bounded linear operators Tn : X Y from a Banach space X into a normed space Y such that ( || Tn || ) is bounded for every xX, say || Tnx || ≤ cx ..………….(1) where cx is a real number. Then the sequence of the norms || Tn || is bounded, that is, there is c such that || Tn || ≤ c …………………………………….(2) 7 Proof. For every kN, let Ak X be the set of all x such that || Tnx || ≤ k for __ all n. To show that Ak is closed, let x Ak , then there is a sequence (xj ) of elements in Ak such that lim xj = x. Then for any fixed n, we have || Tnxj || ≤ k, hence by the continuity of Tn and the continuity of the norm we have, || Tnx || ≤ k and so xAk. Therefore, Ak is closed. Since for every xX there exists cx such that || Tnx || ≤ cx for all n. Then each xX belongs to some Ak. Hence X = Ak , however, X is complete, then by Bairs Category Theorem there exists k 1 __ Ak0 contains an open ball, say B0 = B(x0; r ) Ak0 = Ak0 . Now let xX be r 2|| x|| arbitrary and nonzero and let z = x0 + γx, γ = ≠ 0 ………………(3) r Then || z - x0 || = || γx || = 2 < r. Hence zB0 and so z Ak0. Then by the definition of Ak0 we have, || Tnz || ≤ k0 for all n. Since x0B0 Ak0, then 1 || Tnx0|| ≤ k0 for all n. By (3), x = (z - x0). Then for all n, || Tnx || = 1 1 1 || Tn(z - x0) || ≤ ( || Tnz || + || Tnx0 || ) ≤ (2k0) = 4k0 r || Tn || ≤ . Therefore, || Tn || ≤ c for all n, where c = 2|| x|| r 4k0 r (2k0). Hence 4.7-4 Space of Polynomials. The normed space X of all polynomials is not complete under the norm defined by || x || = max | j | , where α0, α1, α2, … are j the coefficients of x. Proof. We construct a sequence of bounded linear functionals on X which satisfies (1) but not (2). Let x be a nonzero polynomial of degree Nx, then x(t) = t j 0 j j , where αj = 0 for j > Nx. Let ( fn ) be a sequence of functionals that are defined on X by fn(0) = 0, fn(x) = α0 + α1 + α2 … + αn-1. It is left to the reader to show that for any fixed n, fn is linear. Since |αj |≤ max | j | = || x ||, j then | fn(x) | ≤ n|| x ||. Hence fn is bounded. Furthermore, for each fixed xX, | fn(x) | ≤ (Nx + 1) max | j | = cx, (since x is a polynomial of degree Nx that has j Nx + 1 coefficients ). Therefore, (| fn(x) |) satisfies (1). Finally, we show that n (fn) does not satisfy (2). Choose a polynomial x defined by x(t) = t j . Then j 0 | f n ( x )| || x || = 1 and fn(x) = 1 + 1 + … + 1 = n = n || x ||. Hence || fn || ≥ || x|| = n. Hence the sequence (|| fn ||) is unbounded and so (2) is not satisfied. Therefore, by Uniform Boundedness Theorem, X is not complete H. W. 1, 3, 9, 5, 11, 13. H.W.*1 4. 8 4.8 Strong and Weak Convergence 4.8-1 Definition. A sequence ( xn ) in a normed space X is said to be strongly convergent (or convergent in the norm) if there is xX such that || xn – x || 0 as n ∞. This is written as lim xn = x or xn x. x is called the strong limit of ( xn ) and we say that ( xn ) converges strongly to x. 4.8-2 Definition. A sequence ( xn ) in a normed space X is said to be weakly w convergent (xn x) if there is xX such that for every fX /, lim f(xn ) = f(x). x is called the weak limit of ( xn ) and we say that ( xn ) converges weakly to x. 4.8-3 Lemma. Let ( xn ) be a weakly convergent sequence in a normed space w X, say xn x. Then (a) The weak limit x of ( xn ) is unique. (b) Every subsequence of ( xn ) converges weakly to x. (c) The sequence ( || xn || ) is bounded. Proof. The proof of (a) and (b) are left to the reader. w (c) Since xn x, then f(xn ) f(x) for all fX /, so that the sequence of numbers ( f(xn ) ) is bounded. That is, there is cf a constant depending on f such that | f (xn) | ≤ cf for all n. Using the canonical mapping C: X X //, we can define gnX // by gn(f) = f(xn ) for all f X /. Then for all n, | gn(f) | = | f(xn) | ≤ cf , that is the sequence (| gn(f) |) is bounded for all fX/. However, X/ is complete, Then by uniform bounded Theorem 4.7-3, ( || gn || ) is bounded, and by Lemma 4.6-1, || gn || = || xn ||. Hence ( || xn || ) is bounded 4.8-4 Theorem. Let ( xn ) be a sequence in a normed space X. Then (a) Strong convergence implies weak convergence with the same limit. (b) The converse of (a) is not generally true. (c) If dim X < ∞, then weak convergence implies strong convergence. Proof. The proof of (a) is left to the reader. (b) Let (en ) be an orthonormal sequence in a Hilbert space H. By Riez’s representation Theorem, any fH / has a representation f(x) = < x, z >. Hence f(en) = <en, z >. By Bessel’s inequality | en , z |2 ≤ || z || 2. Hence n 1 | e , z | 2 n 1 n converges. So that |<en, z >| 0 as n∞. This implies that f(en) = <en, z>0 = 2 w f(0) as n ∞. Since fH/ was arbitrary, then en 0. However, (en) does not converge strongly, because for all m ≠ n, || em - en ||2 = <em, em > + <en, en> = 2 w (c) Suppose that xn x and dim X = k. Let { e1, e2, …, ek } be any basis of X k and, say, xn = (j n ) e j and x = j 1 k e j 1 j j . By assumption, f(xn ) f(x) for all fX /, in particular fj(xn ) fj(x) for all j = 1, 2, …, k, where fj(ej) = 1 and fj(em) = 0 ( m ≠ j ). Hence fj(xn ) = αj (n) and fj(x) = αj and so αj (n) αj. Then 9 || xn – x || = || k ( (jn) j )e j || ≤ j 1 k | j 1 (n) j j | || ej || 0 as n ∞. Therefore, (xn ) converges strongly to x Examples w 4.8-5 Hilbert Space. In a Hilbert space H, xn x if and only if <xn, z > < x, z > for all z in H. Proof. Left to the reader. p w p 4.8-6 Space . In , where 1 < p < ∞, xn x if and only if: (A) The sequence ( || xn || ) is bounded. (B) For every fixed j we have j (n) j as n ∞; here, xn = (j (n)) and x = (j). Proof. Left to the reader. 4.8-7 Lemma ( Weak Convergence ). In a normed space X we have, w xn x if and only if: (A) The sequence ( || xn || ) is bounded. (B) For every element f of a total subset M of X/ we have, f(xn) f(x) as n∞. w Proof. If xn x, then (A) follows from Lemma 4.8-3, and (B) follows from the definition of weak convergence. Conversely, suppose that (A) and (B) hold. By (A), || xn || ≤ c for all n and || x || ≤ c for sufficiently large c. Since M is total in X/, then for every fX/ there is a sequence (fi) in span M such that fi f (since span M is dense in X / ).Then for every > 0 there is j such that || fi - f || < . Moreover, since fispan M, then by (B) there is N such that | fi(xn) - fi(x) | 3c < 3 for all n > N. Then for all n > N we have, | f(xn) - f(x) | ≤ | f(xn) - fi(xn) | + | fi(xn) - fi(x) | + | f i(x) - f(x) | < || f - fi || || xn || + + 3 + 3c 3 + || fi - f || || x || < 3c c w c = . Since fX/ was arbitrary, then xn x H. W. 1-10 H.W.* 4, 5. 4.9 Convergence of sequences of operators and functionals 4.9-1 Definition. Let X and Y be normed spaces. A sequence ( Tn ) of operators in B(X, Y) is said to be (1) Uniformly convergent if ( Tn ) converges in the norm of B(X, Y), that is there is T :XY such that || Tn – T || 0 as n ∞. (2) Strongly convergent if ( Tnx) converges strongly in Y for every xX, that is there is T :XY such that || Tnx – Tx || 0 as n ∞ for every xX. (3) Weakly convergent if ( Tnx) converges weakly in Y for every xX, that is there is T :XY such that || f(Tnx) – f(Tx) || 0 as n ∞ for every xX and for every fY /. 10 Remark. Uniform operator convergence implies strongly operator convergence implies weakly operator convergence. Proof. Left to the reader. Examples 2 4.9-2 Space . The strong operator convergence need not imply the uniform operator convergence. 2 2 Proof. Let ( Tn ) be a sequence of operators, where Tn: is defined by Tnx = ( 0, 0, ….., 0, n+1, n+2, ……. ), ( note: the number of zeros is n ) where x 2 = (j) . It is easy to see that Tn is linear for any n. Also, for any n, since ||Tnx|| = ( | j |2 ) ≤ ( | j |2 ) = || x ||, then Tn is bounded and || Tn || ≤ 1. 1 2 j n 1 1 2 j 1 Moreover, if we choose x0 = ( 0, 0, ….., 0, n+1, n+2, ……. ) we get, Tnx0 = x0 and so || Tn || ≥ ||Tn x0 || || x0 || = 1. Therefore, || Tn || = 1 and so || Tn – 0 || = || Tn || = 1, which implies that ( Tn) is not uniformly convergent to the zero operator. But, it 2 is clear that for all x we have, Tnx( 0, 0, …..) as n ∞. Hence ( Tn) is strongly operator convergent to 0 2 4.9-3 Space . The weak operator convergence need not imply the strong operator convergence. Proof. Let ( Tn ) be a sequence of operators, where Tn: 2 2 is defined by 2 Tnx = ( 0, 0,..., 0, 1, 2, ... ), where the number of zeros is n and x = (j) It is easy to see that Tn is linear and bounded for any n (how?) Since Hilbert space, then every linear functional f on f(x) = < x, z > = j 1 = k 1 | k 1 |2 j j , where z = (j) 2 . is a has a Riesz representation . But f(Tnx) = <Tnx, z > = j n 1 __ j n j __ k n k . By Cauchy-Schwarz inequality, | f(Tnx) | 2 = | <Tnx, z > | 2 ≤ k __ 2 2 | m n 1 m |2 0 as n ∞. Therefore, f(Tnx) 0 = f(0x) for all x 2 . Thus (Tn) is weakly operator convergent to 0. But, (Tn) is not strongly operator convergent to 0, because for x = (1, 0, 0,...), we have || Tmx–Tnx || = 2 for all m ≠ n Note. Since the real and the complex fields are finite dimensional, then by Theorem 4.8-4 (c), the weak and the strong convergence of any sequence of bounded linear operators are the same. 4.9-4 Definition. Let ( fn ) be a sequence of bounded linear functionals on a normed space X, then (1) strongly convergence of ( fn) means that there is fX/ such that || fn – f || 0 as n ∞. This written as fn f 11 (2) weak* convergence of ( fn) means that there is fX/ such that fn(x) f(x) w as n ∞ for every xX. This written as fn f Remark. Let X and Y be normed spaces, and ( Tn ) be a sequence of elements in B(X, Y). (1) If ( Tn ) converges uniformly to T, then TB(X, Y). (2) Strongly or weakly convergent of ( Tn ) to some T :XY need not imply that TB(X, Y). Proof. The easy proof of (1) is left to the reader. To verify (2), consider the 2 2 subspace X = {x = (j) : x has finitely many nonzero} of . Let Tn:XX be defined by Tnx = (1, 22, 33, ..... , nn, n+1, n+2, ……. ) and T:XX be defined by Tx = (jj). It is left to the reader to show that for any fixed n, Tn is linear and bounded. Moreover, T is linear but is not bounded, then TB(X). Finally, it is clear that ( Tn ) converges strongly to T, where for all xX, || Tnx – Tx || 0 as n ∞ 4.9-5 Lemma. Let X be a Banach space,Y a normed space and ( Tn ) be a sequence of elements in B(X, Y) that converges strongly to a limit T :XY, then TB(X, Y). Proof. It is left to the reader to prove that T is linear. Since Tnx Tx for all xX, then by Lemma 1.4-2(a), the sequence (Tnx) is bounded. However, X is complete, then by U. B. Theorem 4.7-3, there is c > 0 such that || Tn || ≤ c for all n. Hence, || Tnx || ≤ || Tn || || x || ≤ c || x ||. Then by using the continuity of the norm we have, || Tx || = lim || Tnx || ≤ c || x ||. Hence T is bounded and then TB(X, Y) 4.9-6 Theorem. Let X and Y be Banach spaces and ( Tn ) be a sequence of elements in B(X, Y). Then, ( Tn ) converges strongly if and only if: (A) The sequence ( || Tn || ) is bounded. (B) The sequence (Tnx) is Cauchy in Y for every x in a total subset M of X. Proof. If Tnx Tx for all xX, then (Tnx) is bounded and so by U. B. Theorem 4.7-3, ( || Tn|| ) is bounded. (B) follows from the definition of strongly operator convergent and the fact that every convergent sequence is Cauchy. Conversely, suppose that (A) and (B) hold, then there is c > 0 such that ||Tn|| ≤ c for all n. Since M is total in X, then span M is dense in X. Hence for every xX and every > 0 there is yspan M such that || x – y || < 3c . By (B), the sequence (Tny) is Cauchy. Hence there is kN such that || Tny – Tmy || < 3 for m, n > k. Therefore, for all m, n > k and any fixed xX we have, || Tnx – Tmx || ≤ || Tnx – Tny || + || Tny – Tmy || + || Tmy – Tmx || < || Tn|| || x – y || + 3 + || Tm|| || y – x || < c 3c + 3 + c 3c = . Hence (Tnx) is Cauchy in Y for all xX. However, Y is complete, then (Tnx) converges in Y; that is ( Tn ) converges strongly 12 4.9-7 Corollary. A sequence ( fn ) of bounded linear functional on a Banach space X is weak* convergent, the limit being a bounded linear functional on X if and only if: (A) The sequence ( || fn || ) is bounded. (B) The sequence (fnx) is Cauchy for every x in a total subset M of X. H. W. 1-4 H.W.* 4. 4.12 Open Mapping Theorem 4.12-1 Definition. Let X and Y be metric spaces. Then T : D(T) Y with D(T) X is called an open mapping if the image of any open set in D(T) is an open set in Y. 4.12-2 Lemma. A bounded linear operator T from a Banach space X onto a Banach space Y has the property that the image T(B) of the open unit ball B = B(0; 1) X contains an open ball a bout 0Y. 4.12-3 Open Mapping Theorem, Bounded Inverse Theorem. A bounded linear operator from a Banach space X onto a Banach space Y is an open mapping. Hence if T is bijective, then T -1 is continuous and thus is bounded. Proof. Let A be an open set in X. Let y = Tx be an element in T(A). Since A is open, then A contains an open ball with center x. Hence A-x contains an open ball with 1 center 0; let the radius of the ball be r, and set k = r . Then k(A-x) contains the open unit ball B(0; 1). Then by Lemma 4.12.2, T(k(A-x)) = k[T(A) – T(x)] contains a ball a bout 0Y, and so does T(A) – T(x). Hence T(A) cotains an open ball a bout Tx = y. But y was arbitrary in T(A), then T(A) is open. Therefore, T is an open mapping. Finally, if T is bijective, then T -1 exists and linear by Theorem 2.6-10. However, T is open then by Theorem 1.3-4 T is continuous, and by Theorem 2.7-9, T is bounded H. W. 1, 2, 5-7 H.W.* 6. 13 4.13 Closed Linear Operators. Closed Graph Theorem 4.13-1 Definition. Let X and Y be normed spaces and T : D(T) Y be a linear operator with D(T) X. Then T is called a closed linear operator if its graph G(T) = { (x, y) : xD(T), y = Tx } is closed in the normed space XY, where the two algebraic operations of the vector space XY are defined as (x1, y1) + (x2, y2) = (x1+ x2, y1+ y2); (x, y) = (x, y) ( is scalar) and the norm on XY is defined by || (x, y) || = || x || + || y ||. 4.13-2 Closed Graph Theorem. Let X and Y be Banach spaces and let T : D(T) Y be a closed linear operator with D(T) X. If D(T) is closed in X, then T is bounded. Proof. Let (zn) be any Cauchy sequence in XY, where zn = (xn, yn). Then for every > 0 there is kN such that for all m > n > k, || xn – xm || + || yn – ym || = || ( xn – xm, yn – ym )|| = || zn – zm || < ……………………………………...(1) Hence (xn) and (yn) are Cauchy sequences in X and Y, respectively. However, X and Y are complete, then there are xX and yY such that xnx and yn y. Let m∞ in (1) to get, || zn – (x, y) || = || xn – x || + || yn – y || < for all n > k. This means that (zn ) converges to z = (x, y)XY. Therefore, XY is complete. Since G(T) is closed in XY and D(T) is closed in X, then G(T) and D(T) are complete. Now consider the mapping P : G(T) D(T), that defined by P((x, Tx)) = x. It is left to the reader to show that P is linear. Since, || P((x, Tx)) || = || x || for all xX, then P is bounded. It is easy to see that P is bijective (how?) Hence P-1 : D(T) G(T) exists and is defined by P-1(x) = (x, Tx). Since G(T) and D(T) are complete, then by open mapping Theorem 4.12-2, P -1 is bounded, say || (x, Tx) || = || P-1(x) || ≤ b || x || for some b > 0 and all xD(T). Hence || Tx || ≤ || Tx || + || x || = || (x, Tx) || ≤ b || x ||. Therefore, T is bounded. 4.13-3 Theorem. Let T:D(T)Y be a linear operator with D(T)X and X and Y be normed spaces. Then, T is closed if and only if it has the following property: if xnx, where xnD(T) for all n and Txny, then xD(T) and Tx = y………(2) Proof. Suppose that T is closed, then G(T) is closed. Let xnx, where xnD(T) for all n and Txny, then (xn, Txn)G(T) and (xn, Txn)(x, y). Hence (x, y)G(T). This means that xD(T) and Tx = y. Conversely, suppose that the _______ property (2) holds. Let (x, y) G (T ) . Then there is ( (xn, Txn) ) a sequence of elements in G(T) such that (xn, Txn)(x, y). Hence xnx, xnD(T) for all n and Txny. Then by assumption [ the property (2) ], xD(T) and Tx = y; that is (x, y)G(T). Therefore, G(T) is closed, and so T is closed. 4.13-4 Example Defferential operator. Let X = C[0, 1] and T:D(T)X be such that Tx = x /, where x / is the derivative of x and D(T) is a subspace of functions xX which have a continuous derivative. Then T is not bounded, but is closed. 14 Proof. By 2.7-5, T is linear but not bounded. Let xnD(T) for all n such that xnx and Txn = xn / y. By 1.5-5, the convergence in the normed space C[0, t 1] is uniform convergence. Then y ( )d 0 t t = lim x ( )d = lim xn/ ( )d = lim 0 n / n 0 t xn(t) - xn(0) = x(t) - x(0); that is x(t) = x(0) + y ( )d . Then xD(T) and Tx 0 / = x = y. Then by Theorem 4.13-3, T is closed. Note. D(T) in Example 4.13-4, is not closed , because if it is closed then by Theorem 4.13-2, T is bounded, because T is closed. Then we have a contradiction with T is not bounded. Therefore, D(T) in Example 4.13-4, is not closed. Remark. (1) Closedness does not imply boundedness of linear operator. [ see Example 4.13-4 above. ] (2) Boundedness does not imply closedness of linear operator. Proof. (2) Let T : D(T) D(T) be the identity operator on D(T), where D(T) is a proper dense subspace of a normed space X. It is clear that T is linear and bounded. But T is not closed. This follows from Theorem 4.13-3 if we take an x X – D(T) and a sequence (xn) in D(T) which converges to x. 4.13-5 Lemma. Let T : D(T) Y be a bounded linear operator with D(T) X, where X and Y are normed spaces. Then, (a) If D(T) is a closed subset of X, then T is closed. (b) If T is closed and Y is complete, then D(T) is a closed subset of X. Proof. Left to the reader. H. W. 3-6, 11-13. H. W*. 6, 12. 15