Chapter 3.10

Banach Algebras

[26/0%]
Show
LaTeX
Problem 3.10.1

Use the identity (xy)n=x(yx)n−1y(x y)^{n} = x(y x)^{n-1} y to prove that xyx y and yxy x always have the same spectral radius.

?
Problem 3.10.2
?
(a)

If xx and xyx y are invertible in AA, prove that yy is invertible.

(b)

If xyx y and yxy x are invertible in AA, prove that xx and yy are invertible. (The commutative case of this was tacitly used in the proofs of Theorems 10.13 and 10.28).

(c)

Show that it is possible to have xy=ex y = e but yx≠ey x \neq e. For example, consider the right and left shifts SRS_{R} and SLS_{L}, defined on some Banach space of functions ff on the nonnegative integers by

(SRf)(n)=f(n−1) if n≥1(SRf)(0)=0(SLf)(n)=f(n+1) for all n≥0 \begin{aligned} & \left(S_{R} f\right)(n) = f(n-1) \quad \text{ if } n \geq 1 \\ & \left(S_{R} f\right)(0) = 0 \\ & \left(S_{L} f\right)(n) = f(n+1) \quad \text{ for all } n \geq 0 \end{aligned}
(d)

If xy=ex y = e and yx=z≠ey x = z \neq e, show that zz is a nontrivial idempotent. (This means that z2=z,z≠0,z≠ez^{2} = z, z \neq 0, z \neq e).

Problem 3.10.3

Prove that every finite-dimensional AA is isomorphic to an algebra of matrices. Hint: The proof of Theorem 10.2 shows that every AA is isomorphic to a subalgebra of P(A)\mathscr {P}(A). Conclude that xy=ex y = e implies yx=ey x = e if dim⁡A<∞\operatorname {dim} A < \infty.

?
Problem 3.10.4
?
(a)

Prove that e−yxe-y x is invertible in AA if e−xye-x y is invertible. Suggestion: Put z=(e−xy)−1z = (e-x y)^{-1}, write zz as a geometric series (assume ∥x∥<1,∥y∥<1\left\| x\right\| < 1, \left\| y\right\| < 1), and use the identity stated in Exercise 1 to obtain a finite formula for (e−yx)−1(e-y x)^{-1} in terms of x,y,zx, y, z. Then show that this formula works without any restrictions on ∥x∥\left\| x\right\| or ∥y∥\left\| y\right\|.

(b)

If λ∈C,λ≠0\lambda \in \mathcal{C}, \lambda \neq 0, and λ∈σ(xy)\lambda \in \sigma (x y), prove that λ∈σ(yx)\lambda \in \sigma (y x). Thus σ(xy)∪{0}=σ(yx)∪{0}\sigma (x y) \cup \left\{ 0\right\} = \sigma (y x) \cup \left\{ 0\right\}. Show, however, that σ(xy)\sigma (x y) is not always equal to σ(yx)\sigma (y x).

Problem 3.10.5

Let A0A_{0} and A1A_{1} be the algebras of all complex 2-by-2 matrices of the form

[α00β],[αβ0α] \left[\begin{smallmatrix} \alpha & 0 \\ 0 & \beta \end{smallmatrix}\right], \quad \left[\begin{smallmatrix} \alpha & \beta \\ 0 & \alpha \end{smallmatrix}\right]

Prove that every two-dimensional complex algebra AA with unit ee is isomorphic to one of these and that A0A_{0} is not isomorphic to A1A_{1}. Hint: Show that AA has a basis {e,a}\left\{ e, a\right\} in which a2=λea^{2} = \lambda e for some λ∈C\lambda \in \mathbb {C}. Distinguish between the cases λ=0,λ≠0\lambda = 0, \lambda \neq 0. Show that there exists a three-dimensional noncommutative Banach algebra.

?
Problem 3.10.6

Let AA be the algebra of all complex functions ff on {1,2,3,…}\left\{ 1,2,3, \ldots \right\} which are 0 except at finitely many points, with pointwise addition and multiplication and norm

∥f∥=∑k=1∞k−2∣f(k)∣ \left\| f\right\| = \sum _{k = 1}^{\infty } k^{-2}\left|f(k)\right|

Show that multiplication is left-continuous (hence also right-continuous, since AA is commutative) but not jointly continuous. (Adjunction of a unit element, as suggested in Section 10.1, would make no difference). Show, in fact, that there is a sequence {fn}\left\{ f_{n}\right\} in AA so that ∥fn∥→0\left\| f_{n}\right\| \rightarrow 0 but ∥fn2∥→∞\left\| f_{n}^{2}\right\| \rightarrow \infty, as n→∞n \rightarrow \infty.

?
Problem 3.10.7

Let C2=C2([0,1])C^{2} = C^{2}([0,1]) be the space of all complex functions on [0, 1] whose second derivative is continuous. Choose a>0,b>0a > 0, b > 0, and define

∥f∥=∥f∥∞+a∥f′∥∞+b∥f′′∥∞ \left\| f\right\| = \left\| f\right\| _{\infty }+a\left\| f^{\prime }\right\| _{\infty }+b\left\| f^{\prime \prime }\right\| _{\infty }

Show that this makes C2C^{2} into a Banach space, for every choice of a,ba, b, but that the Banach algebra axioms hold if and only if 2b≤a22 b \leq a^{2}. (For necessity, consider xx and x2x^{2}).

?
Problem 3.10.8

What happens if the process of adjoining a unit (described in Section 10.1) is applied to an algebra AA which already has a unit? Clearly, the result cannot be an algebra A1A_{1} with two units. Explain.

?
Problem 3.10.9

Suppose that Ω\Omega is open in CC and that f:Ω→Af: \Omega \rightarrow A and φ:Ω→C\varphi : \Omega \rightarrow C are holomorphic. Prove that φf:Ω→A\varphi f: \Omega \rightarrow A is holomorphic.

?
Remark.
?

This was used in the proof of Theorem 10.13, with φ(λ)=λn\varphi (\lambda ) = \lambda^{n}.

Problem 3.10.10

Another proof that σ(x)≠∅\sigma (x) \neq \varnothing can be based on Liouville's theorem 3.32 and the fact that (λe−x)−1→0(\lambda e-x)^{-1} \rightarrow 0 as λ→∞\lambda \rightarrow \infty. Complete the details.

?
Problem 3.10.11

Call x∈Ax \in A a topological divisor of zero if there is a sequence {yn}\left\{ y_{n}\right\} in AA, with ∥yn∥=1\left\| y_{n}\right\| = 1, such that

lim⁡n→∞xyn=0=lim⁡n→∞ynx \lim _{n \rightarrow \infty } x y_{n} = 0 = \lim _{n \rightarrow \infty } y_{n} x
?
(a)

Prove that every boundary point xx of the set of all invertible elements of AA is a topological divisor of zero. Hint: Take yn=xn−1/∥xn−1∥y_{n} = x_{n}^{-1} /\left\| x_{n}^{-1}\right\|, where xn→xx_{n} \rightarrow x.

(b)

In which Banach algebras is 0 the only topological divisor of 0?

Problem 3.10.12

Find the spectrum of the operator T∈B(ℓ2)T \in \mathscr {B}\left(\ell^{2}\right) given by

T(x1,x2,x3,x4,…)=(−x2,x1,−x4,x3,…) T\left(x_{1}, x_{2}, x_{3}, x_{4}, \ldots \right) = \left(-x_{2}, x_{1},-x_{4}, x_{3}, \ldots \right)
?
Problem 3.10.13

Suppose K={λ∈C:1≤∣λ∣≤2}K = \left\{ \lambda \in \mathscr {C}: 1 \leq \left|\lambda \right| \leq 2\right\}. Put f(λ)=λf(\lambda ) = \lambda. Let AA be the smallest closed subalgebra of C(K)C(K) that contains 1 and ff. Let BB be the smallest closed subalgebra of C(K)C(K) that contains ff and 1/f1 / f. Describe the spectra σA(f)\sigma_{A}(f) and σB(f)\sigma_{B}(f).

?
Problem 3.10.14
?
(a)

Fubini's theorem was applied to vector-valued integrals in the proof of Theorem 10.29. Justify this.

(b)

Construct a proof of Theorem 10.29 that uses no contour integrals, as follows: Prove the theorem first for polynomials gg, then for rational functions gg with no poles in Ω1\Omega_{1}, and obtain the general case from Runge's theorem.

Problem 3.10.15

Suppose XX is a Banach space, T∈B(X)T \in \mathscr {B}(X) is compact, and ∥Tn∥≥1\left\| T^{n}\right\| \geq 1 for all n≥1n \geq 1. Prove that the point spectrum of TT is not empty.

?
Problem 3.10.16

Let X=C([0,1])X = C([0,1]) and define T∈B(X)T \in \mathscr {B}(X) by

(Tx)(s)=∫0sx(u)du(0≤s≤1) (T x)(s) = \int _{0}^{s} x(u) d u \quad (0 \leq s \leq 1)

Show that σp(T)=∅\sigma_{p}(T) = \varnothing. Hence f(σp(T))=∅f\left(\sigma_{p}(T)\right) = \varnothing for every ff, but if f≡0f \equiv 0, then f~(T)=\tilde{f}(T) = 0 ; hence

σp(f~(T))=σp(0)={0}≠∅ \sigma _{p}(\tilde{f}(T)) = \sigma _{p}(0) = \left\{ 0\right\} \neq \varnothing

The extra hypotheses in (c) and (d) of Theorem 10.33 are thus needed.

?
Problem 3.10.17

Suppose that the spectrum of some x∈Ax \in A is not connected. Prove that AA contains a nontrivial idempotent 2. (This is defined in Exercise 2).

Prove also that A=A0⊕A1A = A_{0} \oplus A_{1}, where

A0={x:zx=0},A1={x:zx=x} A_{0} = \left\{ x: z x = 0\right\} , \quad A_{1} = \left\{ x: z x = x\right\}
?
Problem 3.10.18

Suppose Ω\Omega is open in C,α\mathcal{C}, \alpha is an isolated boundary point of Ω,f:Ω→X\Omega , f: \Omega \rightarrow X is a holomorphic XX-valued function (where XX is some complex Banach space), and there is a smallest nonnegative integer nn such that

∣λ−α∣n∥f(λ)∥ \left|\lambda -\alpha \right|^{n}\left\| f(\lambda )\right\|

is bounded as λ→α\lambda \rightarrow \alpha. If n>0n > 0, then ff is said to have a pole of order nn at α\alpha.

?
(a)

Suppose x∈Ax \in A and (λe−x)−1(\lambda e-x)^{-1} has a pole at every point of σ(x)\sigma (x). Prove that there is a nontrivial polynomial PP such that P(x)=0P(x) = 0.

(b)

As a special case of (a)(a), assume σ(x)={0}\sigma (x) = \left\{ 0\right\} and (λe−x)−1(\lambda e-x)^{-1} has a pole of order nn at 0. Prove that xn=0x^{n} = 0.

Remark.
?

In part (a): this can happen only when σ(x)\sigma (x) is a finite set.

Problem 3.10.19

Let SRS_{R} be the right shift, acting on ℓ2\ell^{2}, as in Exercise 2. Let {cn}\left\{ c_{n}\right\} be a sequence of complex numbers such that cn≠0c_{n} \neq 0 but cn→0c_{n} \rightarrow 0 as n→∞n \rightarrow \infty. Define M∈B(ℓ2)M \in \mathscr {B}\left(\ell^{2}\right) by

(Mf)(n)=cnf(n)(n≥0) (M f)(n) = c_{n} f(n) \quad (n \geq 0)

and define T∈B(ℓ2)T \in \mathscr {B}\left(\ell^{2}\right) by T=MSRT = M S_{R}.

?
(a)

Compute ∥Tm∥\left\| T^{m}\right\|, for m=1,2,3,…m = 1,2,3, \ldots.

(b)

Show that σ(T)={0}\sigma (T) = \left\{ 0\right\}.

(c)

Show that TT has no eigenvalue. (Its point spectrum is therefore empty, although its spectrum consists of a single point!)

(d)

Show that (λI−T)−1(\lambda I-T)^{-1} does not have a pole at 0.

(e)

Show that TT is a compact operator.

Problem 3.10.20

Suppose x∈A,xn∈Ax \in A, x_{n} \in A, and lim⁡xn=x\lim x_{n} = x. Suppose Ω\Omega is an open set in C\mathscr {C} that contains a component of σ(x)\sigma (x). Prove that σ(xn)\sigma \left(x_{n}\right) intersects Ω\Omega for all sufficiently large nn. (This strengthens Theorem 10.20). Hint: If σ(x)⊂Ω∪Ω0\sigma (x) \subset \Omega \cup \Omega_{0}, where Ω0\Omega_{0} is an open set disjoint from Ω\Omega, consider the function ff that is 1 in Ω,0\Omega , 0 in Ω0\Omega_{0}.

?
Problem 3.10.21

Let CRC_{R} be the algebra of all real continuous functions on [0,1][0,1], with the supremum norm. This satisfies all requirements of a Banach algebra, except that the scalars are now real.

?
(a)

If ϕ(f)=∫01f(t)dt\phi (f) = \int_{0}^{1} f(t) d t, then ϕ(1)=1\phi (1) = 1, and ϕ(f)≠0\phi (f) \neq 0 if ff is invertible in CRC_{R}, but ϕ\phi is not multiplicative.

(b)

If GG and G1G_{1} are defined in CRC_{R} as in Theorem 10.34, show that G/G1G / G_{1} is a group of order 2.

The analogues of Theorem 10.9 and (d) of Theorem 10.34 are thus false for real scalars. Exactly where would the proof of (d) of Theorem 10.34 break down?

Problem 3.10.22

Suppose A=C(T)A = C(T), the algebra of all continuous complex functions on the unit circle TT, with the supremum norm. Show that two invertible members of C(T)C(T) are in the same coset of G1G_{1} if and only if they are homotopic mappings of TT into the set of all nonzero complex numbers. Deduce from this that G/G1G / G_{1} is isomorphic to the additive group of the integers. (The notation is as in Theorem 10.34).

?
Problem 3.10.23

Suppose A=M(R)A = M(R), the convolution algebra of all complex Borel measures on the real line; see (e) of Example 10.3. Supply the details in the following proof that G/G1G / G_{1} is uncountable: If α∈R\alpha \in R, let δα\delta_{\alpha } be the unit mass concentrated at α\alpha. Assume δα∈G1\delta_{\alpha } \in G_{1}. Then δα=exp⁡(μα)\delta_{\alpha } = \exp \left(\mu_{\alpha }\right) for some μα∈M(R)\mu_{\alpha } \in M(R); hence, for −∞<t<∞-\infty < t < \infty,

−iαt=μ^α(t)+2kπi -i \alpha t = \hat{\mu }_{\alpha }(t)+2 k \pi i

where kk is an integer. Since μ^α\hat{\mu }_{\alpha } is a bounded function, α=0\alpha = 0. Thus δ0\delta_{0} is the only δα\delta_{\alpha } in G1G_{1}. No coset of G1G_{1} in GG contains therefore more than one δα\delta_{\alpha }.

?
Problem 3.10.24
?
(a)

Prove that AA is commutative if there is a constant M<∞M < \infty such that ∥xy∥≤M∥yx∥\left\| x y\right\| \leq M\left\| y x\right\| for all xx and yy in A. Hint: ∥w−1yw∥≤M∥y∥\left\| w^{-1} y w\right\| \leq M\left\| y\right\| if ww is invertible in AA. Replace ww by exp⁡(λx)\exp (\lambda x), where x∈Ax \in A and λ∈C\lambda \in \mathbb {C}. Continue as in Theorem 12.16.

(b)

Prove that AA is commutative if ∥x2∥=∥x∥2\left\| x^{2}\right\| = \left\| x\right\|^{2} for every x∈Ax \in A. Hint: Show that ∥x∥=ρ(x)\left\| x\right\| = \rho (x). Use Exercise 1 to deduce that ∥w−1yw∥=∥y∥\left\| w^{-1} y w\right\| = \left\| y\right\|. Continue as in (a).

Problem 3.10.25

As regards the invariant subspace problem, described in the introduction to Theorem 10.35, explain why the problem is

?
(a)

trivial in Cn\mathscr {C}^{n},

(b)

different in Rn\mathbb {R}^{n},

(c)

uninteresting if XX is not separable.

How should Lomonosov's theorem be reformulated when X=ΦnX = \Phi^{n}?

Problem 3.10.26

Let SRS_{R} be the right shift on ℓ2\ell^{2}, as in Exercise 2. Prove that 0 is the only compact T∈B(ℓ2)T \in \mathscr {B}\left(\ell^{2}\right) that commutes with SRS_{R}. Hint: If T≠0T \neq 0 then

∥T(SRNx)−T(SRMx)∥ \left\| T\left(S_{R}^{N} x\right)-T\left(S_{R}^{M} x\right)\right\|

does not tend to 0 when N−M→∞N-M \rightarrow \infty.

?