Chapter 3.4

Duality in Banach Spaces

[27/0%]
Show
LaTeX
Problem 3.4.1

Let ϕ\phi be the embedding of XX into X∗∗X^{* *} described in Section 4.5. Let τ\tau be the weak topology of XX, and let σ\sigma be the weak*-topology of X∗∗X^{* *}-the one induced by X∗X^{*}.

?
(a)

Prove that ϕ\phi is a homeomorphism of (X,τ)(X, \tau ) onto a dense subspace of (X∗∗,σ)\left(X^{* *}, \sigma \right).

(b)

If BB is the closed unit ball of XX, prove that ϕ(B)\phi (B) is σ\sigma-dense in the closed unit ball of X∗∗X^{* *}. (Use the Hahn-Banach separation theorem).

(c)

Use (a), (b), and the Banach-Alaoglu theorem to prove that XX is reflexive if and only if BB is weakly compact.

(d)

Deduce from (c) that every norm-closed subspace of a reflexive space XX is reflexive.

(e)

If XX is reflexive and YY is a closed subspace of XX, prove that X/YX / Y is reflexive.

(f)

Prove that XX is reflexive if and only if X∗X^{*} is reflexive. Suggestion: One half follows from (c); for the other half, apply (d) to the subspace ϕ(X)\phi (X) of X∗∗X^{* *}.

Problem 3.4.2

Which of the spaces c0,ℓ1,ℓp,ℓ∞c_{0}, \ell^{1}, \ell^{p}, \ell^{\infty } are reflexive? Prove that every finitedimensional normed space is reflexive. Prove that CC, the supremum-normed space of all complex continuous functions, on the unit interval, is not reflexive.

?
Problem 3.4.3

Prove that a subset EE of B(X,Y)\mathscr {B}(X, Y) is equicontinuous if and only if there exists M<∞M < \infty such that ∥Λ∥≤M\left\| \Lambda \right\| \leq M for every Λ∈E\Lambda \in E.

?
Problem 3.4.4

Recall that X∗=B(X,C)X^{*} = \mathscr {B}(X, \mathscr {C}), if C\mathscr {C} is the scalar field. Hence Λ∗∈B(C,X∗)\Lambda^{*} \in \mathscr {B}\left(\mathbb {C}, X^{*}\right) for every Λ∈X∗\Lambda \in X^{*}. Identify the range of Λ∗\Lambda^{*}.

?
Problem 3.4.5

Prove that T∈B(X,Y)T \in \mathscr {B}(X, Y) is an isometry of XX onto YY if and only if T∗T^{*} is an isometry of Y∗Y^{*} onto X∗X^{*}.

?
Problem 3.4.6

Let σ\sigma and τ\tau be the weak*-topologies of X∗X^{*} and Y∗Y^{*}, respectively, and prove that SS is a continuous linear mapping of (Y∗,τ)\left(Y^{*}, \tau \right) into (X∗,σ)\left(X^{*}, \sigma \right) if and only if S=T∗S = T^{*} for some T∈B(X,Y)T \in \mathscr {B}(X, Y).

?
Problem 3.4.7

Let L1L^{1} be the usual space of integrable functions on the closed unit interval JJ, relative to Lebesgue measure. Suppose T∈B(L1,Y)T \in \mathscr {B}\left(L^{1}, Y\right), so that T∗∈B(Y∗,L∞)T^{*} \in \mathscr {B}\left(Y^{*}, L^{\infty }\right). Suppose R(T∗)\mathscr {R}\left(T^{*}\right) contains every continuous function on JJ. What can you deduce about TT?

?
Problem 3.4.8

Prove that (ST)∗=T∗S∗(S T)^{*} = T^{*} S^{*}. Supply the hypotheses under which this makes sense.

?
Problem 3.4.9

Suppose S∈B(X),T∈B(X)S \in \mathscr {B}(X), T \in \mathscr {B}(X).

?
(a)

Show, by an example, that ST=IS T = I does not imply TS=IT S = I.

(b)

However, assume TT is compact, show that

S(I−T)=I if and only if (I−T)S=I S(I-T) = I \text{ if and only if }(I-T) S = I

and show that either of these equalities implies that I−(I−T)−1I-(I-T)^{-1} is compact.

Problem 3.4.10

Assume T∈B(X)T \in \mathscr {B}(X) is compact, and assume either that dim⁡X=∞\operatorname {dim} X = \infty or that the scalar field is C\mathbb {C}. Prove that σ(T)\sigma (T) is not empty. However, σ(T)\sigma (T) may be empty if dim⁡X<∞\operatorname {dim} X < \infty and the scalar field is R\mathbb {R}.

?
Problem 3.4.11

Suppose dim⁡X<∞\operatorname {dim} X < \infty and show that the equality β∗=β\beta^{*} = \beta of Theorem 4.25 reduces to the statement that the row rank of a square matrix is equal to its column rank.

?
Problem 3.4.12

Suppose T∈B(X,Y)T \in \mathscr {B}(X, Y) and R(T)\mathscr {R}(T) is closed in YY. Prove that

dim⁡N(T)=dim⁡X∗/R(T∗)dim⁡N(T∗)=dim⁡Y/R(T) \begin{aligned} \operatorname {dim} \mathscr {N}(T) & = \operatorname {dim} X^{*} / \mathscr {R}\left(T^{*}\right) \\ \operatorname {dim} \mathscr {N}\left(T^{*}\right) & = \operatorname {dim} Y / \mathscr {R}(T) \end{aligned}

This generalizes the assertions α=β∗\alpha = \beta^{*} and α∗=β\alpha^{*} = \beta of Theorem 4.25.

?
Problem 3.4.13
?
(a)

Suppose T∈B(X,Y),Tn∈B(X,Y)T \in \mathscr {B}(X, Y), T_{n} \in \mathscr {B}(X, Y) for n=1,2,3,…n = 1,2,3, \ldots, each TnT_{n} has finitedimensional range, and lim⁡∥T−Tn∥=0\lim \left\| T-T_{n}\right\| = 0. Prove that TT is compact.

(b)

Assume YY is a Hilbert space, and prove the converse of (a): Every compact T∈B(X,Y)T \in \mathscr {B}(X, Y) can be approximated in the operator norm by operators with finite-dimensional ranges. Hint: In a Hilbert space there are linear projections of norm 1 onto any closed subspace. (See Theorems 5.16, 12.4).

Problem 3.4.14

Define a shift operator SS and a multiplication operator MM on ℓ2\ell^{2} by

(Sx)(n)={0 if n=0x(n−1) if n≥1(Mx)(n)=(n+1)−1x(n) if n≥0 \begin{aligned} (S x)(n) & = \begin{cases} 0 & \text{ if } n = 0 \\ x(n-1) & \text{ if } n \geq 1\end{cases} \\ (M x)(n) & = (n+1)^{-1} x(n) \quad \text{ if } n \geq 0 \end{aligned}

Put T=MST = M S. Show that TT is a compact operator which has no eigenvalue and whose spectrum consists of exactly one point. Compute ∥Tn∥\left\| T^{n}\right\|, for n=1,2,3,…n = 1,2,3, \ldots, and compute lim⁡n→∞∥Tn∥1/n\lim_{n \rightarrow \infty }\left\| T^{n}\right\|^{1 / n}.

?
Problem 3.4.15

Suppose μ\mu is a finite (or σ\sigma-finite) positive measure on a measure space Ω,μ×μ\Omega , \mu \times \mu is the corresponding product measure on Ω×Ω\Omega \times \Omega, and K∈L2(μ×μ)K \in L^{2}(\mu \times \mu ). Define

(Tf)(s)=∫ΩK(s,t)f(t)dμ(t)[f∈L2(μ)] (T f)(s) = \int _{\Omega } K(s, t) f(t) d \mu (t) \quad \left[f \in L^{2}(\mu )\right]
?
(a)

Prove that T∈B(L2(μ))T \in \mathscr {B}\left(L^{2}(\mu )\right) and that

∥T∥2≤∬ΩΩ∣K(s,t)∣2dμ(s)dμ(t) \left\| T\right\| ^{2} \leq \iint _{\Omega \Omega }\left|K(s, t)\right|^{2} d \mu (s) d \mu (t)
(b)

Suppose ai,bia_{i}, b_{i} are members of L2(μ)L^{2}(\mu ), for 1≤i≤n1 \leq i \leq n, put K1(s,t)=∑ai(s)bi(t)K_{1}(s, t) = \sum a_{i}(s) b_{i}(t), and define T1T_{1} in terms of K1K_{1} as TT was defined in terms of KK. Prove that dim⁡R(T1)≤n\operatorname {dim} \mathscr {R}\left(T_{1}\right) \leq n.

(c)

. Deduce that TT is a compact operator on L2(μ)L^{2}(\mu ). Hint: Use Exercise 13.

(d)

Suppose λ∈C,λ≠0\lambda \in \mathbb {C}, \lambda \neq 0. Prove: Either the equation

Tf−λf=g T f-\lambda f = g

has a unique solution f∈L2(μ)f \in L^{2}(\mu ) for every g∈L2(μ)g \in L^{2}(\mu ) or there are infinitely many solutions for some gg and none for others. (This is known as the Fredholm alternative).

(e)

Describe the adjoint of TT.

Problem 3.4.16

Define

K(s,t)={(1−s)t if 0≤t≤s(1−t)s if s≤t≤1 K(s, t) = \begin{cases} (1-s) t & \text{ if } 0 \leq t \leq s \\ (1-t) s & \text{ if } s \leq t \leq 1\end{cases}

and define T∈B(L2(0,1))T \in \mathscr {B}\left(L^{2}(0,1)\right) by

(Tf)(s)=∫01K(s,t)f(t)dt(0≤s≤1) (T f)(s) = \int _{0}^{1} K(s, t) f(t) d t \quad (0 \leq s \leq 1)
?
(a)

Show that the eigenvalues of TT are (nπ)−2,n=1,2,3,…(n \pi )^{-2}, n = 1,2,3, \ldots, that the corresponding eigenfunctions are sin⁡nπx\sin n \pi x, and that each eigenspace is onedimensional. Hint: If λ≠0\lambda \neq 0, the equation Tf=λfT f = \lambda f implies that ff is infinitely differentiable, that λf′′+f=0\lambda f^{\prime \prime }+f = 0, and that f(0)=f(1)=0f(0) = f(1) = 0. The case λ=0\lambda = 0 can be treated separately.

(b)

Show that the above eigenfunctions form an orthogonal basis for L2(0,1)L^{2}(0,1).

(c)

Suppose g(t)=∑cnsin⁡nπtg(t) = \sum c_{n} \sin n \pi t. Discuss the equation Tf−λf=gT f-\lambda f = g.

(d)

Show that TT is also a compact operator on CC, the space of all continuous functions on [0,1]. Hint: If {fi}\left\{ f_{i}\right\} is uniformly bounded, then {Tfi}\left\{ T f_{i}\right\} is equicontinuous.

Problem 3.4.17

If L2=L2(0,∞)L^{2} = L^{2}(0, \infty ) relative to Lebesgue measure, and if

(Tf)(s)=1s∫0sf(t)dt(0<s<∞) (T f)(s) = \frac{1}{s} \int _{0}^{s} f(t) d t \quad (0 < s < \infty )

prove that T∈B(L2)T \in \mathscr {B}\left(L^{2}\right) and that TT is not compact. (The fact that ∥T∥≤2\left\| T\right\| \leq 2 is a special case of Hardy's inequality. See p. 72 of [23]).

?
Problem 3.4.18

Prove the following statements:

?
(a)

If {xn}\left\{ x_{n}\right\} is a weakly convergent sequence in XX, then {∥xn∥}\left\{ \left\| x_{n}\right\| \right\} is bounded.

(b)

If T∈B(X,Y)T \in \mathscr {B}(X, Y) and xn→xx_{n} \rightarrow x weakly, then Txn→TxT x_{n} \rightarrow T x weakly.

(c)

If T∈B(X,Y)T \in \mathscr {B}(X, Y), if xn→xx_{n} \rightarrow x weakly, and if TT is compact, then ∥Txn−Tx∥→0\left\| T x_{n}-T x\right\| \rightarrow 0.

(d)

Conversely, if XX is reflexive, if T∈B(X,Y)T \in \mathscr {B}(X, Y), and if ∥Txn−Tx∥→0\left\| T x_{n}-T x\right\| \rightarrow 0 whenever xn→xx_{n} \rightarrow x weakly, then TT is compact. Hint: Use (c) of Exercise 1, and part (c) of Exercise 28 in Chapter 3.

(e)

If XX is reflexive and T∈B(X,ℓ1)T \in \mathscr {B}\left(X, \ell^{1}\right), then TT is compact. Hence R(T)≠ℓ1\mathscr {R}(T) \neq \ell^{1}. Hint: Use (c) of Exercise 5 of Chapter 3.

(f)

If YY is reflexive and T∈B(c0,Y)T \in \mathscr {B}\left(c_{0}, Y\right), then TT is compact.

Problem 3.4.19

Suppose YY is a closed subspace of XX, and x0∗∈X∗x_{0}^{*} \in X^{*}. Put

μ=sup⁡{∣⟨x,x0∗⟩∣:x∈Y,∥x∥≤1}δ=inf⁡{∥x∗−x0∗∥:x∗∈Y⊥} \begin{aligned} & \mu = \sup \left\{ \left| \left\langle x, x_{0}^{*} \right\rangle \right|: x \in Y, \left\| x\right\| \leq 1\right\} \\ & \delta = \inf \left\{ \left\| x^{*}-x_{0}^{*}\right\| : x^{*} \in Y^{\perp }\right\} \end{aligned}

In other words, μ\mu is the norm of the restriction of x0∗x_{0}^{*} to YY, and δ\delta is the distance from x0∗x_{0}^{*} to the annihilator of YY. Prove that μ=δ\mu = \delta. Prove also that δ=∥x∗−x0∗∥\delta = \left\| x^{*}-x_{0}^{*}\right\| for at least one x∗∈Y⊥x^{*} \in Y^{\perp }.

?
Problem 3.4.20

Extend Sections 4.6 to 4.9 to locally convex spaces. (The word "isometric" must of course be deleted from the statement of Theorem 4.9).

?
Problem 3.4.21

Let BB and B∗B^{*} be the closed unit balls in XX and X∗X^{*}, respectively. The following is a converse of the Banach-Alaoglu theorem: If EE is a convex set in X∗X^{*} such that E∩(rB∗)E \cap \left(r B^{*}\right) is weak*-compact for every r>0r > 0, then EE is weak*-closed. (Corollary: A subspace of X∗X^{*} is weak*-closed if and only if its intersection with B∗B^{*} is weak*-compact).

Complete the following outline of the proof.

(i) EE is norm-closed.

(ii) Associated to each F⊂XF \subset X its polar

P(F)={x∗:∣⟨x,x∗⟩∣≤1 for all x∈F} P(F) = \left\{ x^{*}:\left| \left\langle x, x^{*} \right\rangle \right| \leq 1 \text{ for all } x \in F\right\}

The intersection of all sets P(F)P(F), as FF ranges over the collection of all finite subsets of r−1Br^{-1} B, is exactly rB∗r B^{*}.

(iii) The theorem is a consequence of the following proposition: If, in addition to the stated hypotheses, E∩B∗=∅E \cap B^{*} = \varnothing, then there exists x∈Xx \in X such that Re(⟨x,x∗⟩)≥1\mathrm{Re}\left( \left\langle x, x^{*} \right\rangle \right) \geq 1 for every x∗∈Ex^{*} \in E.

(iv) Proof of the proposition: Put F0={0}F_{0} = \left\{ 0\right\}. Assume finite sets F0,…,Fk−1F_{0}, \ldots , F_{k-1} have been chosen so that iFi⊂Bi F_{i} \subset B and so that

P(F0)∩⋯∩P(Fk−1)∩E∩kB∗=∅(1) P\left(F_{0}\right) \cap \cdots \cap P\left(F_{k-1}\right) \cap E \cap k B^{*} = \varnothing \tag {1}

Note that (1) is true for k=1k = 1. Put

Q=P(F0)∩⋯∩P(Fk−1)∩E∩(k+1)B∗ Q = P\left(F_{0}\right) \cap \cdots \cap P\left(F_{k-1}\right) \cap E \cap (k+1) B^{*}

If P(F)∩Q≠∅P(F) \cap Q \neq \varnothing for every finite set F⊂k−1BF \subset k^{-1} B, the weak*-compactness of QQ, together with (ii), implies that (kB∗)∩Q≠∅\left(k B^{*}\right) \cap Q \neq \varnothing, which contradicts (1). Hence there is a finite set Fk⊂k−1BF_{k} \subset k^{-1} B such that (1) holds with k+1k+1 in place of kk. The construction can thus proceed. It yields

E∩⋂k=1∞P(Fk)=∅(2) E \cap \bigcap _{k = 1}^{\infty } P\left(F_{k}\right) = \varnothing \tag {2}

Arrange the members of ⋃Fk\bigcup F_{k} in a sequence {xn}\left\{ x_{n}\right\}. Then ∥xn∥→0\left\| x_{n}\right\| \rightarrow 0. Define T:X∗→c0T: X^{*} \rightarrow c_{0} by

Tx∗={⟨xn,x∗⟩} T x^{*} = \left\{ \left\langle x_{n}, x^{*} \right\rangle \right\}

Then T(E)T(E) is a convex subset of c0c_{0}. By (2),

∥Tx∗∥=sup⁡n∣⟨xn,x∗⟩∣≥1 \left\| T x^{*}\right\| = \sup _{n}\left| \left\langle x_{n}, x^{*} \right\rangle \right| \geq 1

for every x∗∈Ex^{*} \in E. Hence there is a scalar sequence {αn}\left\{ \alpha_{n}\right\}, with ∑∣αn∣<∞\sum \left|\alpha_{n}\right| < \infty, such that

Re(∑n=1∞αn⟨xn,x∗⟩)≤1 \mathrm{Re}\left(\sum _{n = 1}^{\infty } \alpha _{n} \left\langle x_{n}, x^{*} \right\rangle \right) \leq 1

for every x∗∈Ex^{*} \in E. To complete the proof, put x=∑αnxnx = \sum \alpha_{n} x_{n}.

?
Problem 3.4.22

Suppose T∈B(X),TT \in \mathscr {B}(X), T is compact, λ≠0\lambda \neq 0, and S=T−λIS = T-\lambda I.

?
(a)

If N(Sn)=N(Sn+1)\mathscr {N}\left(S^{n}\right) = \mathscr {N}\left(S^{n+1}\right) for some nonnegative integer nn, prove that N(Sn)=N(Sn+k)\mathscr {N}\left(S^{n}\right) = \mathscr {N}\left(S^{n+k}\right) for k=1,2,3,…k = 1,2,3, \ldots

(b)

Prove that (a) must happen for some n. (Hint: Consider the proof of Theorem 4.24).

(c)

Let nn be the smallest nonnegative integer for which (a) holds. Prove that dim⁡N(Sn)\operatorname {dim} \mathscr {N}\left(S^{n}\right) is finite, that

X=N(Sn)⊕R(Sn) X = \mathscr {N}\left(S^{n}\right) \oplus \mathscr {R}\left(S^{n}\right)

and that the restriction of SS to R(Sn)\mathscr {R}\left(S^{n}\right) is a one-to-one mapping of R(Sn)\mathscr {R}\left(S^{n}\right) onto R(Sn)\mathscr {R}\left(S^{n}\right).

Problem 3.4.23

Suppose {xn}\left\{ x_{n}\right\} is a sequence in a Banach space XX, and

∑n=1∞∥xn∥=M<∞ \sum _{n = 1}^{\infty }\left\| x_{n}\right\| = M < \infty

Prove that the series ∑xn\sum x_{n} converges to some x∈Xx \in X. Explicitly, prove that

lim⁡n→∞∥x−(x1+⋯+xn)∥=0 \lim _{n \rightarrow \infty }\left\| x-\left(x_{1}+\cdots +x_{n}\right)\right\| = 0

Prove also that ∥x∥≤M\left\| x\right\| \leq M. (These facts were used in the proof of Theorem 4.13).

?
Problem 3.4.24

Let cc be the space of all complex sequences

x={x1,x2,x3,…} x = \left\{ x_{1}, x_{2}, x_{3}, \ldots \right\}

for which x∞=lim⁡xnx_{\infty } = \lim x_{n} exists (in C\mathbb {C}). Put ∥x∥=sup⁡∣xn∣\left\| x\right\| = \sup \left|x_{n}\right|. Let c0c_{0} be the subspace of cc that consists of all xx with x∞=0x_{\infty } = 0.

?
(a)

Describe explicitly two isometric isomorphisms uu and vv, such that uu maps c∗c^{*} onto ℓ1\ell^{1} and vv maps c0∗c_{0}^{*} onto ℓ1\ell^{1}.

(b)

Define SS : c0→cc_{0} \rightarrow c by Sf=fS f = f. Describe the operator vS∗u−1v S^{*} u^{-1} that maps ℓ1\ell^{1} to ℓ1\ell^{1}.

(c)

Define T:c→c0T: c \rightarrow c_{0} by setting

y1=x∞,yn+1=xn−x∞ if n≥1 y_{1} = x_{\infty }, \quad y_{n+1} = x_{n}-x_{\infty } \quad \text{ if } n \geq 1

Prove that TT is one-to-one and that Tc=c0T c = c_{0}. Find ∥T∥\left\| T\right\| and ∥T−1∥\left\| T^{-1}\right\|. Describe the operator uT∗v−1u T^{*} v^{-1} that maps ℓ1\ell^{1} to ℓ1\ell^{1}.

Problem 3.4.25

If T∈B(X,Y)T \in \mathscr {B}(X, Y) and R(T∗)=N(T)⊥\mathscr {R}\left(T^{*}\right) = \mathscr {N}(T)^{\perp }, prove that R(T)\mathscr {R}(T) is closed.

?
Problem 3.4.26

Assume T∈B(X,Y)T \in \mathscr {B}(X, Y) and T(X)=YT(X) = Y. Show that there exists δ>0\delta > 0 such that S(X)=YS(X) = Y for all S∈B(X,Y)S \in \mathscr {B}(X, Y) with ∥S−T∥<δ\left\| S-T\right\| < \delta.

?
Problem 3.4.27

Suppose T∈B(X)T \in \mathscr {B}(X). Prove that λ∈σ(T)\lambda \in \sigma (T) if and only if there is a sequence {xn}\left\{ x_{n}\right\} in X,∥xn∥=1X,\left\| x_{n}\right\| = 1, for which

lim⁡n→∞∥Txn−λxn∥=0 \lim _{n \rightarrow \infty }\left\| T x_{n}-\lambda x_{n}\right\| = 0
?
Remark.
?

Thus every λ∈σ(T)\lambda \in \sigma (T) which is not an eigenvalue of TT is an "approximate" eigenvalue.