Mathematics 442C Suggested solutions to exercise sheet 5 1. Let A and B be Banach algebras, and let A0 be a dense subalgebra of A. If θ0 : A0 → B is a continuous homomorphism, show that there is a unique continuous homomorphism θ : A → B such that θ|A0 = θ0 . Solution. We first prove something slightly different, which you may have seen before: if θ0 : A0 → B is a continuous linear map between a vector space A0 which is dense in a Banach space A, and a Banach space B, then there is a unique continuous linear map θ : A → B such that θ|A0 = θ0 . For uniqueness, recall that any two continuous mappings A → B which agree on a dense subset of A must be equal (in fact, this is true for any topological spaces A and B). For existence, if a ∈ A, let θ(a) = limn→∞ θ0 (an ) where an is any sequence in A0 which converges to a. To check that this all makes sense, observe that • since A0 is dense in A, such a sequence (an ) exists for every a ∈ A; • since θ0 is continuous, it maps convergent sequences in A to Cauchy sequences in B, which are convergent since B is complete. So the limit limn→∞ θ0 (an ) above exists; • if an → a and bn → a as n → ∞ where an , bn ∈ A0 , then an − bn → 0 ∈ A0 , so θ0 (an − bn ) = θ0 (an ) − θ0 (bn ) → 0, so limn→∞ θ0 (an ) = limn→∞ θ0 (bn ). So the definition of θ(a) above does not depend on the choice of convergent sequence (an ); • if a ∈ A0 then by choosing the constant sequence an = a we see that θ(a) = limn→∞ θ0 (a) = θ0 (a), so θ|A0 = θ0 ; • if a, b ∈ A and λ ∈ C and an , bn ∈ A0 with an → a and bn → b as n → ∞, then an + λbn → a + λb, so θ(a + λb) = limn→∞ θ0 (an + λbn ) = limn→∞ θ0 (an ) + λ limn→∞ θ0 (bn ) = θ(a) + λθ(b). So θ is linear; and • θ0 is a continuous linear map, hence is bounded, and if A0 3 an → a ∈ A then kθ(a)k = k limn→∞ θ0 (an )k = limn→∞ kθ0 (an )k ≤ kθ0 k limn→∞ kan k = kθ0 k kak, so kθk = kθ0 k. Hence θ is a bounded linear map, so is continuous. Now consider the situation in the original question where A0 is a dense subalgebra of A and θ0 is a continuous homomorphism. We wish to show that the continuous linear map θ constructed above is actually a homomorphism. If a, b ∈ A then there are sequences (an )n≥1 and (bn )n≥1 in A0 such that an → a and bn → b as n → ∞. Since multiplication is jointly continuous in a Banach algebra, we also have an bn → ab as n → ∞. Since θ is continuous and θ0 is a homomorphism, we have θ(ab) = lim θ0 (an bn ) by the definition of θ n→∞ = lim θ0 (an )θ0 (bn ) since θ0 is a homomorphism n→∞ = lim θ0 (an ) lim θ0 (bm ) since multiplication is jointly continuous in B n→∞ m→∞ = θ(a)θ(b). So θ is a homomorphism. 2. Let a be a normal element of a unital C*-algebra. Prove the following assertions. (a) a is hermitian if and only σ(a) ⊆ R. (b) a is unitary if and only if σ(a) ⊆ T. (c) a is a projection if and only if σ(a) ⊆ {0, 1}. If a is not normal, then these assertions may all fail. Find an example to demonstrate this. Solution. (a) If a is hermitian then σ(a) ⊆ R by Corollary 4.1.10. Conversely, suppose that a is normal and σ(a) ⊆ R. Let z : σ(a) → C, t → 7 t. ∗ ∗ Observe that z ∈ C(σ(a)) and z = z since z (t) = z(t) = t = t for t ∈ σ(a). The continuous functional calculus at a is a ∗-homomorphism with z(a) = a, so ∗ a∗ = z(a) = (z ∗ )(a) = z(a) = a. So a is hermitian. (b) If a is unitary then σ(a) ⊆ T by Corollary 4.1.10. Conversely, suppose that a is normal and σ(a) ⊆ T. Let z : σ(a) → C, λ 7→ λ. Observe that z ∈ C(σ(a)) and zz ∗ = z ∗ z = 1, since (z ∗ z)(λ) = λλ = |λ|2 = 1 = 1(λ) for λ ∈ σ(a). The continuous functional calculus at a is a unital ∗-homomorphism with z(a) = a, so a∗ a = z(a)∗ z(a) = z ∗ (a)z(a) = (zz ∗ )(a) = 1(a) = 1 and similarly, aa∗ = 1. So a is unitary. (c) If a is a projection then a2 = a, so a2 − a = 0. If λ ∈ σ(a) then λ2 − λ ∈ σ(a2 − a) = {0} by the polynomial spectral mapping theorem. So λ(λ − 1) = 0, so λ ∈ {0, 1}. Conversely, suppose that a is normal and σ(a) ⊆ {0, 1}. We know that a is hermitian, by (a). Let z : σ(a) → C, t 7→ t. Observe that z ∈ C(σ(a)) and z 2 = z since z 2 (t) = z(t)2 = t2 = t = z(t) for t ∈ σ(a). The continuous functional calculus at a is a ∗homomorphism with z(a) = a, so a2 = z(a2 ) = z(a)2 = (z 2 )(a) = z(a) = a. So a∗ = a = a2 , so a is a projection. To see that the assertions may fail if a is not normal, let A = M2 (C) and let a = 10 11 . Then σ(a) = {1} = R ∩ T ∩ {0, 1}, but a is not normal. So a is neither hermitian, unitary nor a projection. 3. Let A be a unital C*-algebra. For a ∈ A, recall that ea was defined in Exercise 2.9 as ea = lim pn (a) where pn (x) = n→∞ n X xk k=0 k! Show that if h is a hermitian element of A, then eih is unitary. 2 . ∗ Solution. Since h = h∗ , we have (ih)k = (−ih)k . Hence pn (ih)∗ = pn (−ih). Since the adjoint is continuous, this gives ∗ (eih )∗ = lim pn (ih) = lim pn (ih)∗ = lim pn (−ih) = e−ih . n→∞ n→∞ n→∞ Since ih clearly commutes with −ih, Exercise 2.9(f) gives (eih )∗ eih = e−ih eih = e−ih+ih = e0 = 1, and similarly, eih (eih )∗ = eih e−ih = eih−ih = e0 = 1. So eih is unitary. Another way to do this question is to show that if a is normal, then ea = exp(a) where exp ∈ C(σ(a)) is the restriction of the exponential function C → C to σ(a), and then use the spectral mapping theorem and the fact that exp(iR) = T and Exercise 2. 4. Prove Lemma 4.3.5: Let A be a unital C*-algebra. If a is a normal element of A then C ∗ (1, a) is a unital abelian C*-algebra. Solution. By the definition of C ∗ (1, a), it is a C*-subalgebra containing 1. Hence it is a unital C*-algebra, with unit 1. By Lemma 4.3.4 and using the normality of a (aa∗ = a∗ a) we have that C ∗ (1, a) is the closure of the linear span of S = {1, an (a∗ )m : n, m ≥ 1}. If x, y ∈ S then xy = yx, since a is normal and 1 commutes with any element of A. Hence any two elements of the linear span of S commute. Taking limits and using the continuity of the product in A, it follows that any two elements of the closure of the linear span of S commute. Hence C ∗ (1, a) is abelian. 5. Prove Lemma 4.3.8: Suppose that X and Y are compact Hausdorff topological spaces and ψ : X → Y is a homeomorphism. The map ψ t : C(Y ) → C(X), f 7→ f ◦ ψ is a well-defined isometric unital ∗-isomorphism. Solution. Since ψ is continuous, if f ∈ C(Y ) then f ◦ ψ = ψ t (f ) is a continuous map X → C. So ψ t is well-defined. Moreover, for x ∈ X we have ψ t (1C(Y ) )(x) = 1C(Y ) (ψ(x)) = 1 = 1C(X) (x), so ψ t (1C(Y ) ) = 1C(X) and ψ t is unital. For f, g ∈ C(Y ) and λ ∈ C we have: ψ t (f + λg)(x) = (f + λg)(ψ(x)) = f (ψ(x)) + λg(ψ(x)) = (ψ t (f ) + λψ t (g))(x), ψ t (f g)(x) = (f g)(ψ(x)) = f (ψ(x)) · g(ψ(x)) = (ψ t (f ) · ψ t (g))(x), ψ t (f ∗ )(x) = f ∗ (ψ(x)) = f (ψ(x)) = ψ t (f )(x) = (ψ t (f )∗ )(x). 3 So ψ t is a ∗-homomorphism. Moreover, since ψ is surjective, we have kψ t (f )k = sup |f (ψ(x))| = sup |f (y)| = kf k x∈X y∈Y so ψ t is an isometry and in particular, ψ t is injective. For h ∈ C(X) we have h = h ◦ ψ −1 ◦ ψ = ψ t (h ◦ ψ −1 ). Since h ◦ ψ −1 ∈ C(X), this shows that ψ t is surjective. 6. Let A be a unital C*-algebra. (a) Show that if a, b ∈ A+ with ab = ba then ab ∈ A+ . (b) Give an example with A = M2 (C) to show that there may be a, b ∈ A+ such that ab 6∈ A+ . Solution. (a) Let B = C ∗ ({1, a, b}), a unital abelian C*-subalgebra of A. Since the Gelfand representation γ : B → C(Ω(B)) for B is a unital ∗-isomorphism, it is spectrumpreserving and sends hermitian elements to hermitian elements. So γ(B + ) = C(Ω(B))+ . Since a, b ∈ A+ we have a, b ∈ B + , so γ(a), γ(b) ∈ C(Ω(B))+ . The (pointwise) product of two positive-valued continuous functions is positive-valued, so γ(a)γ(b) = γ(ab) ∈ C(Ω(B))+ . Since γ −1 is a ∗-isomorphism, it also preserves positivity, so ab = γ −1 (γ(ab)) ∈ B + ⊆ A+ . 1 i 1 0 (b) For example, consider a, b ∈ M2 (C) given by a = 21 −i 1 and b = 0 0 . Then it is easy to check that a and b are hermitian with a2 = a and b2 = b, so they are both ∗ 1 0 1 i projections. In particular, they are positive. However, ab = ( −i 0 ) 6= (ab) = ( 0 0 ), so ab is not hermitian. So ab is not positive. 7. Prove Lemma 4.4.9: If A is a unital C*-algebra and a ∈ A+ then a ≤ kak1. Solution. We have r(a) ≤ kak and a is hermitian, so σ(a) ⊆ [−kak, kak]. Hence σ(kak1 − a) = kak − σ(a) ⊆ kak − [−kak, kak] = [0, 2kak] ⊆ R+ . Since kak1 − a is hermitian, this shows that it is positive. Hence kak1 − a ≥ 0, so a ≤ kak1. 8. Let A be a unital C*-algebra. Show that S(A) is a convex subset of A∗ . Solution. We have S(A) = {τ ∈ A∗ : kτ k = τ (1A ) = 1} = {τ ∈ A∗ : kτ k ≤ 1, τ (1A ) = 1}. Let τ1 , τ2 ∈ S(A) and let t ∈ [0, 1], and let τ = tτ1 + (1 − t)τ2 . Since A∗ is a vector space we have τ ∈ A∗ . Moreover, τ (1A ) = tτ1 (1A ) + (1 − t)τ2 (1A ) = t + 1 − t = 1, and by the triangle inequality, kτ k ≤ tkτ1 k + (1 − t)kτ2 k = 1. So τ ∈ S(A). Hence S(A) is convex. 4 9. Let A be a unital abelian Banach algebra. Let a be a normal element of A and let f ∈ C(σ(a)) and g ∈ C(σ(f (a))). (a) Check that f (a), (g ◦ f )(a) and g(f (a)) are well-defined elements of A. Explain why it is not obvious that (g ◦ f )(a) = g(f (a)). (b) Show that (g ◦ f )(a) = g(f (a)) if g is of the following special forms: (i). g(λ) = λk for some integer k ≥ 0; (ii). g(λ) = λk λj for some integers k, j ≥ 0; (iii). g is a trigonometric polynomial, meaning that g(λ) = n X αk,j λk λj , λ ∈ σ(f (a)) k,j=0 for some αk,j ∈ C. (c) Use the Stone-Weierstrass theorem to deduce that (g ◦ f )(a) = g(f (a)) for any g ∈ C(σ(f (a))). [Hint: approximate g.] Solution. [Whoops, I didn’t actually want A to be abelian in this question. So I won’t assume that below]. (a) Since f ∈ C(σ(a)) and a is normal, f (a) = θa (f ) is well-defined by the continuous functional calculus. The domain of g is σ(f (a)) = f (σ(a)) by the spectral mapping theorem. So the domain of g is the range of f , and g ◦ f is a well-defined function. It is the composition of two continuous functions, so g ◦ f is continuous; so g ◦ f ∈ C(σ(a)). Since a is normal, (g ◦ f )(a) = θa (g ◦ f ) is well-defined by the continuous functional calculus. The element f (a) is in the abelian C*-algebra C ∗ (1, a), and any element of an abelian C*-algebra is normal (since its adjoint is in the algebra, so commutes with it). Hence f (a) is normal, and g ∈ C(σ(f (a))). So g(f (a)) = θf (a) (g) is well-defined by the continuous functional calculus. The way in which (g ◦ f )(a) = θa (g ◦ f ) and g(f (a)) = θf (a) (g) are defined, using the functional calculus based at two different elements (a and f (a)) of A makes it unclear why they might be equal. (b) (i) If g(λ) = λk then g ◦ f = f k (the pointwise kth power of f ) so, using the homomorphism property of the functional calculus (and the fact that it is unital, for k = 0) gives (g ◦ f )(a) = (f k )(a) = f (a)k = g(f (a)). (ii) We have g = g1 g2 = g1 g2∗ where g1 , g2 are of the form considered in (i). So g ◦ f = (g1 g2 ∗ ) ◦ f = (g1 ◦ f )(g2∗ ◦ f ) = (g1 ◦ f )(g2 ◦ f )∗ . Since the functional calculus is a ∗-homomorphism and using (i) we have (g ◦ f )(a) = (g1 ◦ f )(a) (g2 ◦ f )(a)∗ = g1 (f (a)) g2 (f (a))∗ = g1 (f (a)) (g2∗ )(f (a)) = (g1 g2∗ )(f (a)) = g(f (a)). 5 (iii) This is immediate from (ii), since the functional calculi at a and at f (a) are both linear. (c) By the Stone-Weierstrass theorem, the set of trigonometric polynomials is dense in C(σ(f (a)); indeed, it is a unital ∗-subalgebra containing the function λ 7→ λ, so it separates the points of σ(f (a)). Let g ∈ C(σ(f (a))). If ε > 0 then, by the previous paragraph, there is a trigonometric polynomial h ∈ C(σ(f (a)) with kg − hk < ε. Then kg ◦ f − h ◦ f k = k(g − h) ◦ f k < ε in the norm of C(σ(a)) [Why?]. By (c) we have h(f (a)) = (h ◦ f )(a). Now the continuous functional calculus is an isometry, so kg(f (a)) − h(f (a))k = k(g − h)(f (a))k = kθf (a) (g − h)k = kg − hk < ε and k(g ◦ f )(a) − (h ◦ f )(a)k = k((g − h) ◦ f )(a)k = kθa ((g − h) ◦ f )k = k(g − h) ◦ f k < ε. By the triangle inequality, kg(f (a)) − (g ◦ f )(a)k < 2ε and ε was an arbitrary positive number. Hence g(f (a)) = (g ◦ f )(a). 10. Let I be an index set, and for each i ∈ I, let Hi be a Hilbert space and let Ti ∈ B(Hi ) with supi∈I kTi k < ∞. Show that the operator M M T = Ti ∈ B Hi i∈I i∈I satisfies kT k = supi∈I kTi k. [All norms in the statement of this question are operator norms.] L Solution. Let H = i∈I Hi (the Hilbertian direct sum). Let M = supi∈I kTi k. Then M = sup{kTi hi k : i ∈ I, hi ∈ Hi , khi k ≤ 1} so if kT k < M then there is i ∈ I and a vector hi ∈ Hi with khi k ≤ 1 such that kT k < kTi hi k. Let x = (xj )j∈I be given by xj = 0 if j 6= i, and xi = hi . Clearly, x ∈ H and kxk = khi k ≤ 1, and if y = T x then yj = 0 if j 6= i and yi = Ti hi . Hence X kTi hi k2 > kT k2 ≥ kT xk2 = kyk2 = kyj k2 = kTi hi k2 , j∈I which is a contradiction. So kT k ≥ M . Conversely, let h ∈ H with khk ≤ 1. We have X X X kT hk2 = kTi hi k2 ≤ kTi k2 khi k2 ≤ sup kTi k2 khi k2 = M 2 khk2 . i∈I i∈I i∈I So kT k ≤ M . By the previous paragraph, we have equality. 6 i∈I 11. (a) Let H be a Hilbert space. A unit vector is a vector x ∈ H with kxk = 1. Given a unit vector x, let ωx : B(H) → C, T 7→ hT x, xi. Show that ωx is a state on B(H) for every unit vector x. (b) Let A be a unital C*-algebra and let τ ∈ S(A), and consider the Hilbert space H and the ∗-homomorphism π : A → B(H) constructed in Lemma 4.5.9. Find a unit vector x ∈ H such that τ (a) = hπ(a)x, xi for a ∈ A. (c) The maps ωx considered in (a) are called vector states. If A is a unital C*-algebra, consider the Hilbert space H and the isometric ∗-isomorphism π : A → B(H) constructed in the GNS Theorem 4.5.10. We call π the universal representation of A. Show that “every state is a vector state in the universal representation”; or, more precisely, show that S(A) = {ωy ◦ π : y is a unit vector in H}. [Warning: H and π are different in parts (b) and (c)!] Solution. (a) Let ω = ωx where x is a unit vector. We have ω(S + λT ) = h(S + λT )x, xi = hSx, xi + λhT x, xi = ω(S) + λω(T ) so ω is a linear functional; and by the Cauchy-Schwarz inequality, |ω(T )| = |hT x, xi| ≤ kT xk kxk ≤ kT k so ω ∈ B(H)∗ with kωk ≤ 1; and ω(I) = hIx, xi = hx, xi = kxk2 = 1. So ω ∈ B(H)∗ with kωk = ω(I) = 1, so ω is a state on B(H). (b) Using the notation of the proof of Lemma 4.5.9, let x = [1A ]. Then x ∈ H and kxk = h[1A ], [1A ]i = (1A , 1A ) = τ (1∗A 1A ) = τ (1A ) = 1. So x is a unit vector, and 2 hπ(a)x, xi = hπ(a)[1A ], [1A ]i = h[a1A ], [1A ]i = (a, 1A ) = τ (a). (c) If y is a unit vector in H then τ = ωy ◦ π : A → C satisfies kτ k = kωy ◦ πk ≤ kωy k kπk = 1 and τ (1A ) = ωy (π(1A )) = ωy (I) = 1 so τ ∈ S(A). Conversely, let τ ∈ S(A), and let Hτ and πτ be the corresponding Hilbert space and ∗-homomorphism fromLLemma 4.5.9. By (b), τ (a) = hπτ (a)x, xi for some unit vector x ∈ Hτ . Let y ∈ H = σ∈S(A) Hσ be the vector with yσ = 0 if σ 6= τ and yτ = x. Then y is a unit vector, and X (ωy ◦ π)(a) = hπ(a)y, yi = hπσ (a)yσ , yσ i = hπτ (a)x, xi = τ (a). σ∈S(A) 7