1 Basics of Algebras

1.1 Let ϕ be a non-zero linear funional on an algebra A over complex numbers, satisfying ⟨ϕ, ab⟩ = ⟨ϕ, a⟩⟨ϕ, b⟩.
Such a linear funional ϕ is also called a complex homomorphism. Show that

(1) if A has identity e then ϕ(e) = 1;

(2) for any invertible a ∈ A , it holds that ϕ(a) ̸= 0.

Proof. (1) Since ϕ is nonzero, there exists a such that ϕ(a) ̸= 0. en ϕ(a) = ϕ(ae) = ϕ(a)ϕ(e), and thus
ϕ(e) = 1.
(2) 1 = ϕ(e) = ϕ(aa−1 ) = ϕ(a)ϕ(a−1 ).

1.2 Let J be an ideal of algebra A . Show that J is maximal iff A /J does not contain a non-zero ideal.

Proof. `Only if ': Let J be a maximal ideal of A . Suppose that B = A /J contains a non-zero ideal JB . Consider
the natural maps ϕ : A → B and ψ : B → B/JB . It is clear that both ϕ and ψ are non-trivial homomorphisms,
and thus ker(ψ ◦ ϕ) is an ideal of A containing J. Since JB is non-zero, there exists a ∈ A such that [a] ∈ JB
and [a] ̸= [0]. erefore a ∈ ker(ψ · ϕ) but a ̸∈ J, which contradis with the maximality of J.
`If ': Suppose B = A /J does not contain a non-zero ideal but J ′ ⊃ J is a bigger ideal of A . Consider JB =
{[x] ∈ B : x ∈ J ′ \ J}. Since J ′ \ J ̸= ∅, we know that JB ̸= ∅. From J ′ ̸= A we also know that JB ̸= B.
Lastly, for all [a] ∈ B and [j] ∈ JB , it holds that [a][j] = [aj] = [j] and [j][a] = [ja] = [j] because ja ∈ J ′ and
aj ∈ J ′ as J ′ is an ideal. We have found that JB is a non-zero ideal of B. Contradiion.

2 Banach Algebra
2.1 Let A be a Banach algebra with identity and G(A ) be the set of all invertible elements in A . Show that G(A ) is
open and a 7→ a−1 is continuous.

Proof. We shall use the next problem in the proof. Let a ∈ G(A ). en for all b ∈ B(a, ∥a−1
1
∥ ), we have that
−1 −1 −1
∥a (b − a)∥ < 1. Hence e + a (b − a) is invertible and b = a(e + a (b − a)) is invertible.
To show the continuity of inverse map, we first observe that for a, b ∈ G(A ) it holds that

∥b−1 − a−1 ∥ = ∥(a + η)−1 − a−1 ∥

= ∥(a(e + a−1 η))−1 − a−1 ∥
= ∥((e + a−1 η)−1 − e)a−1 ∥
≤ ∥(e + a−1 η)−1 − e∥ ∥a−1 ∥
∥a−1 η∥
≤ ∥a−1 ∥
1 − ∥a−1 η∥
∥a−1 ∥ ∥η∥
≤ ∥a−1 ∥
1 − ∥a−1 ∥ ∥η∥

For a given ϵ > 0 we can choose δ such that

δ∥a−1 ∥2
< ϵ,
1 − δ∥a−1 ∥

then ∥b−1 − a−1 ∥ < ϵ whenever ∥b − a∥ < δ.

1
2.2 Let A be a Banach algebra with identity and a ∈ A with ∥a∥ < 1. Show that e − a ∈ G(A ) and
∞
∑
(e − a)−1 = an .
n=0

∑N
Proof. Let yN = n=0 an . Since ∥a∥ < 1, it is easy to show that {yN } is Cauchy and hence yN converges to some
y ∈ A . Observe that (e − a)yN = yN (e − a) = e − aN +1 → e as N → ∞. Since multiplication is continuous,
we have that (e − a)y = y(e − a) = e, which shows that e − a ∈ G(A ) and (e − a)−1 = y.

2.3 Let A be a Banach algebra with identity and a ∈ ∂(G(A )). Prove that
(1) If an ∈ G(A ), an → a, then limn→∞ ∥a−1
n ∥ = ∞.
(2) ere exists bn ∈ A , ∥bn ∥ = 1, such that limn→∞ abn = limn→∞ bn a = 0.

Proof. (1) If ∥a−1 −1 −1

n ∥ ≤ L for all n, then ∥an (a − an )∥ < 1 for n sufficiently large, which means that e + an (a −
−1
an ) ∈ G(A ) and a = an (e + an (a − an )) ∈ G(A ). Contradiion.
(2) Suppose that an → a with an ∈ G(A ). Let bn = a−1 −1
n /∥an ∥, then

∥aa−1
n ∥
∥abn ∥ =
∥a−1
n ∥
∥(a − an + an )a−1
n ∥
= −1
∥an ∥
∥(a − an )a−1
n + e∥
= −1
∥an ∥
∥(a − an )a−1
n ∥ + ∥e∥
≤
∥a−1
n ∥
∥e∥
≤ ∥a − an ∥ + −1 → 0
∥an ∥

as n → ∞. Hence abn → 0. Similarly we can show that bn a → 0.

2.4 Let {( ) }
α β
A = : α, β ∈ C
0 α
be an algebra under the usual addition and multiplication of matrices. Show that A is a Banach algebra under the
norm ( )
α β

0 α = |α| + |β|.

Proof. e only less trivial part is to show the completeness. Suppose that {An } ⊆ A is a Cauchy sequence. Since
∥An − Am ∥ = |αn − αm | + |βn − βm |, we know that {αn } and {αm }(are Cauchy ) sequences, hence αn → α and
α β
βn → β for some α and β. It is then straightforward to see that An → .
0 α

2.5 Let A be a Banach algebra with identity and ϕ : A → C a homomorphism. en |ϕ(a)| ≤ ∥a∥ for all a ∈ A .

Proof. Suppose that |ϕ(a)| > ∥a∥ for some a. Let b = a/|ϕ(a)|, then |ϕ(b)| = 1 > ∥b∥, and e − b is invertible.
us 1 = ϕ(e) = ϕ(e − b)ϕ((e − b)−1 ). Note that ϕ(e − b) = ϕ(e) − ϕ(b) = 0. Contradiion.

2
 2.6 Let A be a commutative Banach algebra with identity. Show that a ∈ A is invertible if and only if ϕ(a) ̸= 0 for
all non-trivial continuous homomorphism ϕ : A → C.

Proof. `Only if ': Trivial, as 1 = ϕ(e) = ϕ(a)ϕ(a−1 ).

`If ': Suppose that a is not invertible. Let J be a maximal ideal containing a (J exists because aA is an ideal). en
by Gelfand-Mazur eorem, A /J is isomorphic to C, and there exists a natural continuous homomorphism from
ϕJ : A → C such that ϕJ (a) = 0. Contradiion.

2.7 Let A be a Banach algebra with identity and a ∈ A . Show that

(1) σ(a) is compa;
(2) σ(a) is not empty.

Proof. (1) Let λ ∈ σ(s). It is clear that |λ| ≥ ∥a∥, otherwise λe − a = λ(e − a/λ) would be invertible as
∥a/λ∥ < 1. Hence σ(a) is bounded.
Now we show that ρ(a) is open. Suppose that λ ∈ ρ(a), then λe − a ∈ G(A ). Recall that G(A ) is open,
hence there exists ϵ such that λe − a + η ∈ G(A) whenever ∥η∥ < ϵ. In particular, choose η = δe, where
|δ| < ϵ/∥e∥, it follows that (λ + δ)e − a is invertible.
(2) Suppose that σ(a) = ∅, then λe − a is invertible for all λ. Define r(λ) = (λe − a)−1 . e rest of the proof
goes as in the proof of Gelfand-Mazur eorem. Finally we arrive at r(λ) = 0 for all λ. Contradiion.

2.8 Let A be a Banach algebra with identity and a, b ∈ A . Show that

(1) If e − ab is invertible then e − ba is invertible, too;
(2) If λ ∈ σ(ab), λ ̸= 0, then λ ∈ σ(ba);
(3) If a is invertible then σ(ab) = σ(ba).

Proof. (1) (e − ba)−1 = e + b(e − ab)−1 a.

(2) λe − ba is invertible =⇒ e − λ−1 ba is invertible =⇒ (by the first part) e − λ−1 ab is invertible =⇒ λe − ab is
invertible. Contradiion.
(3) λe − ab is invertible ⇐⇒ λa−1 − b is invertible ⇐⇒ λe − ba is invertible.

2.9 Let A and B be commutative Banach algebras with identity and B semi-simple. Let ϕ : A → B be a homomor-
phism, show that ϕ is continuous.

Proof. We shall show that ϕ is closed, whence the continuity follows from Closed Graph eorem. Suppose that
an → 0 in A and ϕ(an ) → b in B. We want to show that b = 0. If not, since B is semi-simple, there exists a
maximal ideal J such that b ̸∈ J. By Gelfand-Mazur eorem, there exists an isomorphism i : B/J → C, and
i([b]) ̸= 0. Note that i ◦ ψ ◦ ϕ : A → C is a homomorphism, where ψ is the natural homomorphism from B to
B/J. By Problem 2.4, |i ◦ ψ ◦ ϕ(an )| ≤ ∥an ∥ for all n. Let n → ∞, we find that |i ◦ ψ(b)| ≤ 0, thus i ◦ ψ(b) = 0
and ψ(b) = 0. Contradiion.

2.10 Let A be a Banach algebra with identity. Let

r(a) = sup{|λ| : λ ∈ σ(a)},

which is called the speral radius of a. Show that for all a, b ∈ A ,

1
(1) r(a) = limn→∞ ∥an ∥ n ;
(2) r(ab) = r(ba);
(3) if ab = ba then r(a + b) ≤ r(a) + r(b) and r(ab) ≤ r(a)r(b).

3
 1
Proof. (1) According to Cauchy-Hadamard test, when λ > lim supn→∞ ∥an ∥ n , it holds that
∑∞
an
(λe − a)−1 = .
n=0
λn+1

erefore r(a) ≤ lim supn→∞ ∥an ∥ n . Next we show that lim supn→∞ ∥an ∥ n ≤ r(a). For all ϕ ∈ A ∗ , the
1 1

funion λ 7→ ⟨ϕ, (λe − a)−1 ⟩ is analytic in the region |λ| > r(a). Hence
∑∞
⟨ϕ, an ⟩
⟨ϕ, (λe − a)−1 ⟩ = .
n=0
λn+1

and ∮
1
⟨ϕ, an ⟩ = ⟨ϕ, (λe − a)−1 ⟩λn dλ
2πi |λ|=r(a)+ϵ

Let M = max|λ|=r(a)+ϵ ∥(λe − a)−1 ∥, then M < ∞ because (λe − a)−1 is continuous w.r.t. λ. us

|⟨ϕ, an ⟩| ≤ ∥ϕ∥M (r(a) + ϵ)n+1

for all ϕ ∈ A ∗ . Hence by Hahn-Banach eorem,

∥an ∥ ≤ M (r(a) + ϵ)n+1 .

erefore
1
lim ∥an ∥ n ≤ r(a) + ϵ.
n→∞
1
Let ϵ → 0 and we conclude that lim supn→∞ ∥an ∥ n = r(a).
1
Now we show that r(a) ≤ lim infn→∞ ∥an ∥ n . Note that

λn e − an = (λe − a)P (a) = P (a)(λe − a),

where
P (a) = λn−1 e + λn−2 a + · · · + an−1 ,
1
whence we see that λ ∈ σ(a) implies that λn ∈ σ(an ). It follows that |λn | ≤ ∥an ∥, and |λ| ≤ lim infn→∞ ∥an ∥ n .
(2) By Exercise 2.8(2), σ(ab) and σ(ba) differ by 0 only at most. If both σ(ab) and σ(ba) contain a non-zero
number, we would have r(ab) = r(ba). Now suppose that σ(ab) = {0}, then λe − ab is invertible for any
λ ̸= 0. By Exercise 2.8(2), λe − ba is invertible, too, and thus σ(ba) = {0} since σ(ba) ̸= ∅.
1 1 1 1
(3) Since ab = ba, it holds that ∥(ab)n ∥ n = ∥an bn ∥ n ≤ ∥an ∥ n ∥bn ∥ n . It follows immediately from part (1)
that r(ab) ≤ r(a)r(b).
Pick α > r(a) and β > r(b), let x = a/α and y = b/β. en
n ( ) n1 ( n ( ) ) n1
∑ n ∑ n
1 k n−k
∥(a + b) ∥
n n = a b ≤ k n−k
α β ∥x ∥ ∥y
k n−k
∥ .
k k
k=0 k=0

Let
kn = arg max ∥xk ∥ ∥y n−k ∥,
k
so
1 1 1
∥(a + b)n ∥ n ≤ ∥xkn ∥ n ∥y n−kn ∥ n (α + β)
for all n. Since 0 ≤ kn /n ≤ 1 we can choose a subsequence such that kni /ni → δ for some δ as i tends to
infinity. Denote this subsequence by kn . If δ = 0 then
1 kn
lim sup ∥xkn ∥ n ≤ lim sup ∥x∥ n ≤ 1,
n→∞ n→∞

4
 otherwise kni ̸= 0 for i big enough and thus

1
( 1
) knn
lim sup ∥xkn ∥ n = lim sup ∥xkn ∥ kn = r(x)δ ≤ 1
n→∞ n→∞

because r(x) ≤ ∥x∥ < 1. erefore r(a + b) ≤ α + β and the conclusion follows by letting α → r(a)+ and
β → r(b)+ .

2.11 Let A = {f ∈ C 1 [0, 1]} with norm

∥f ∥C 1 = ∥f ∥ + ∥f ′ ∥.
Show that A is a semi-simple commutative Banach algebra.

Proof. It is clear that A is a commutative Banach algebra with identity. We shall show the semi-simpleness by
showing
1
lim ∥f n ∥ n = 0 =⇒ f = 0.
n→∞

In fa, r(f ) = 0 means that λe − f is invertible for all λ ̸= 0. If f (x0 ) = λ ̸= 0 for some x0 then λe − f would
not be invertible. Hence f (x) = 0 for all x ∈ [0, 1].

∥a2 ∥ ∥â∥∞
2.12 Let A be a commutative Banach algebra and r = inf and s = inf . Show that s2 ≤ r ≤ s.
a̸=0 ∥a∥ 2 a̸=0 ∥a∥

Proof. Note that ∥Γa∥2 = ∥Γa2 ∥ ≤ ∥a2 ∥, it follows that s2 ≤ r. Now we show the second half. Starting from
∥a2 ∥ ≥ r∥a∥2 , then ∥a4 ∥ ≥ r∥a2 ∥2 ≥ r3 ∥a∥4 . By induion one can show that ∥a2 ∥ ≥ r2 −1 ∥a∥2 , hence
k k k

k 1 1
∥a2 ∥ 2k ≥ r1− 2k ∥a∥

Letting k → ∞, we obtain that ∥â∥∞ ≥ r∥a∥. It follows immediately that s ≥ r.

3 Examples and Applications

3.1 Let { ∞
}
∑
|n|
A = f : Z → C : ∥f ∥ = |f (n)|2 <∞
n=−∞

under the usual addition of scalar multiplication and the following multiplication
∞
∑
f ∗ g(n) = f (n − k)g(k).
k=−∞

Show that
(1) A is a commutative Banach algebra;
(2) Let K = {z ∈ C : 12 ≤ |z| ≤ 2} then K is one-to-one correspondent with M and the Gelfand representation
of A is the Laurent series that are absolutely convergent on K.

Proof. (1) It is easy to verify that A is a commutative algebra, and

∞  ∑
 ∞ 
∑ 
 
∥f ∗ g∥ =  f (n − k)g(k) 2|n|
n=−∞
 
k=−∞

5
 ∞
∑ ∞
∑
≤ |f (n − k)| |g(k)|2|n|
n=−∞ k=−∞
∞
( ∞
)
∑ ∑
= |f (n − k)|2|n| |g(k)|
k=−∞ n=−∞
∑∞
≤ (∥f ∥2|k| )|g(k)|
k=−∞
= ∥f ∥ ∥g∥.
e only thing left is to show that A is complete. Suppose that ∑ {fn } is a Cauchy sequence, i.e., for given
ϵ there exists N such that for all m > n ≥ N it holds that k∈Z |fm (k) − fn (k)|2|k| < ϵ. is implies
∑
that {fn (k)} is Cauchy and thus fn (k) → f (k) for some f . Note that k∈Z |fm (k) − f (k)|2|k| ≤ ϵ, i.e.,
fm − f ∈ A , and therefore f ∈ A .
∑
(2) For f ∈ A define gf (z) = n∈Z f (n)z n , which is well-defined on K. Given z0 ∈ K, consider homo-
morphism ϕz0 : f 7→ gf (z0 ), then Jz0 = ker ϕz0 is a maximal ideal, since gf (z) is continuous. Obviously
z0 7→ Jz0 is injeive, and we shall show that this mapping is surjeive also. Let J ∈ M, we want to find
z0 ∈ K such that ϕJ = ϕz0 , i.e., ⟨ϕJ , f ⟩ = gf (z0 ) for all f ∈ A . Let h ∈ A such that gh (z) = z, then
ghn (z) = z n . Since |⟨ϕJ , h⟩n | = |⟨ϕJ , hn ⟩| ≤ ∥hn ∥ = 2|n| for all n, it follows that ⟨ϕJ , h⟩ ∈ K, say z0 ,
then by continuity of ϕJ , ⟨ϕJ , f ⟩ = gf (z0 ) for all f ∈ A . Aually f 7→ gf is the Gelfand representation of
A.

3.2 Let A be the semi-simple commutative Banach algebra in Problem 2.11. Find its maximal ideal space M. For
x ∈ [0, 1] define
J = {f ∈ A : f (x) = f ′ (x) = 0},
show that J is a closed ideal of A and A /J is a two-dimensional algebra with one-dimensional radical.

Proof. Similarly to eorem 5.3.1, we have that M is homeomorphic and isomorphic to [0, 1]. For a given x ∈ [0, 1],
it is obvious that J is a closed ideal. It is easy to verify that ϕ : A /J → C2 as ϕ(f ) = (f (x), f ′ (x)) is an
isomorphism, where the multiplication of C2 is defined as
(x1 , y1 )(x2 , y2 ) = (x1 x2 , x1 y2 + x2 y1 ).
It is then easy to see that the identity element of C2 is (1, 0) and all the non-invertible elements have the form
(0, z). In fa, those non-invertible elements constitute the only maximal ideal in A . Hence the radical is one-
dimensional.

3.3 Let M be a compa T2 space. Show that there exists a one-to-one correspondence between the set of all closed
subsets of M and the set of closed ideals of C(M ).

Proof. Suppose X is a closed subset of M . Define JX = {f ∈ C(M ) : f (x) = 0 for all x ∈ X}, which is clearly
a closed ideal of C(M ). e map X 7→ JX is clearly ∩injeive by Urysohn's Lemma. Now we prove the converse.
Suppose that J is a closed ideal of C(M ), let X = f ∈J {x : f (x) = 0}, which is the interseion of closed sets,
and thus closed. From eorem 5.3.1 we know that X is non-empty and thus J ⊆ JX . Let f ∈ JX . Given any
positive ϵ let Fϵ = {x ∈ M : |f (x)| ≥ ϵ} then Fϵ is compa and disjoint from X. If there exists gϵ ∈ J such that
gϵ = 1 on Fϵ and |gϵ | ≤ 1 on M , then f gϵ ∈ J and ∥f − f gϵ ∥ ≤ ϵ, which implies that f ∈ I since J is closed.
erefore J = JX .
Now we shall constru such gϵ . For any x ̸∈ Fϵ there exists fx ∈ J such that fx (x) ̸= 0 and thus fx (x) is non-zero
on a neighbhourhood of x. Since Fϵ is compa, we can choose a finite cover of the neighbourhoods corresponding
to x1 , . . . , xn . en
∑n ∑
n
hϵ (x) = fxi (x)fxi (x) = |fxi (x)|2
i=1 i=1

6
 is contained in J and positive on Fϵ . Since Fϵ is continuous, hϵ attains minimum c at some x0 . Now let kϵ (x) =
max{hϵ (x), c}, then kϵ ∈ C(M ), kϵ > 0 everywhere and kϵ = hϵ on Fϵ . Finally let gϵ = kϵ−1 hϵ .

3.4 Let A = C n [0, 1] with norm

∑
n
|f (k) (t)|
∥f ∥ = sup .
0≤t≤1 k!
k=0

Show that, under the usual addition, multiplication and scalar multiplication of funions, A is a Banach algebra.
How to charaerise its maximal ideals?

Proof. It is easy to verify that A is an algebra and ∥f g∥ ≤ ∥f ∥ ∥g∥. To show the completeness of A recall that if
un → u and u′n → v uniformly, where un , u, u′n , v are continuous, then u′ = v. It is then straightforward to see
that A is complete, and thus a Banach algebra. Similar to the case of C 1 [0, 1], the maximal ideals are homomorphic
and isomorphic to [0, 1].

3.5 Define positionwise multiplication on ℓ1 . Show that ℓ1 is a commutative Banach algebra without identity. Further-
more, show that
(1) ere exists a one-to-one correspondence between M and Z;
(2) Gelfand topology is discrete topology;
(3) ere exists a one-to-one correspondence between the set of closed ideals of ℓ1 and the set of subsets of Z.

Proof. It is clear that ℓ1 is commutative. If it has identity e, then e = (1, 1, 1, . . . ), which is not in ℓ1 . Hence ℓ1
has no identity. It is also clear that ∥xy∥ ≤ ∥x∥ ∥y∥. erefore ℓ1 is a Banach algebra. For the next problems, we
can replace Z by Z+ , the set of non-negative integers.

(1) Let n ∈ Z+ . Consider Jn = {x ∈ ℓ1 : xn = 0}, which is clearly an ideal. It is also easy to see that Jn is
maximal, and the map n 7→ Jn is injeive.
For n = 0, consider J0 = {x ∈ ℓ1 : ∃Nx ∀n ≥ Nx , xn = 0}, which is also clearly a maximal ideal.
Now we prove the converse. Let J be a maximal ideal. If for any k there exists xk ∈ J such that xk,k ̸= 0,
then xk,k ek ∈ J and thus ek ∈ J. For any x ∈ ℓ1 , all finite truncations of x are in J. We conclude that
J0 ⊆ J and by maximality of J, we have that J = J0 . Otherwise, there exists n such that xn = 0 for all
x ∈ J, so that J ⊆ Jn and thus J = Jn by maximality of J.
(2) From part (1) it is clear that x̂(n) = xn for all x ∈ ℓ1 . ere exists x ∈ ℓ1 such that xi ̸= xj for all pair i ̸= j.
Since x̂ is continuous under Gelfand topology, it must be discrete.
(3) For any non-empty subset I ⊆ Z+ , consider JI = {x ∈ ℓ1 : xn = 0 for all n ∈ I}. It is clear that JI is a
∩ ideal, and the map I 7→ JI is injeive. We shall show the converse. Let J be a closed ideal. Consider
closed
I = x∈J Z(x) where Z(x) = {n ∈ Z+ : xn = 0}. We claim that I ̸= ∅.
If for any k there exists xk ∈ J such that xk,k ̸= 0, then xk,k ek ∈ J and thus ek ∈ J. As a consequence, for
any x ∈ ℓ1 , all finite truncations of x are in J. Since J is closed, we conclude that J = ℓ1 . Contradiion.
Hence I ̸= ∅.
We have seen that J ⊆ JI . Now we shall show that J = JI . From a similar argument to the above, we
have ek ∈ J for all k ̸∈ I, hence any x ∈ JI can be approximated by its finite truncations, and J = JI . e
conclusion follows from the closedness of J.

3.6 Let A be a semi-simple commutative Banach algebra. Prove that Γ(A ) is closed in C(M) if and only if there exists
a constant K such that ∥a∥2 ≤ K∥a2 ∥ for all a ∈ A .

7
 Proof. We use the notations of Problem 5.2.12.
`If ': e assumptions implies that r ≥ K 1
> 0, thus s > r > 0, i.e., s∥a∥ ≤ ∥â∥ for all a ∈ A . Suppose that
{c
an } is a Cauchy sequence in C(M), then ∥an ∥ is a Cauchy sequence in A . Hence there exists a ∈ A such that
an → a. By continuity of Γ, it holds that ac
n →b a, whence it follows that Γ(A ) is closed.
`Only if ': It suffices to show that s ̸= 0 then r ≥ s2 > 0. e Gelfand transform a 7→ â is a continuous isomorphism
between two Banach spaces, A and C(M), as Γ(A ) is closed and A is semisimple. By Open Mapping eorem,
Γ−1 exists and is continuous, that is, there exists c such that ∥Γ−1 â∥ ≤ c∥â∥ for all â ∈ C(A ), i.e., ∥a∥ ≤ c∥â∥
for all a ∈ A .

4 C ∗ -algebra
4.1 Let A be a commutative Banach algebra. Suppose that A is semi-simple, then every involution on A is continuous.

Proof. By Closed Graph eorem it suffices to show that involution is closed, that is, suppose that am → 0 and
a∗m → a, we need to show that a = 0. Since A is semi-simple, we need only to show that r(a) = 0. In fam
r(a) ≤ r(a − a∗m ) + r(a∗m ) ≤ ∥a − a∗m ∥ + ∥a∗m ∥ → 0.

4.2 Verify that L1 (R) is a Banach algebra under the multiplication of convolution
∫
(x ∗ y)(t) = x(s)y(t − s)ds
R

and involution
x∗ (t) = x(−t).
Is it a C ∗ -algebra?

Proof. We have that

∫ ∫
∗
(x ∗ y) (t) = (x ∗ y)(−t) = x(s)y(−t − s)ds = x(s)y(t − s)ds
R R
∫ ∫
= x(−s)y(−t + s)ds = x∗ (s)y ∗ (t − s)ds = x∗ y ∗ = y ∗ x∗
R R

since convolution is commutative. e rest assumptions are easy to verify. Hence L1 (R) is a Banach algebra with
involution as defined. It is not a C ∗ -algebra. Take f (x) = (sgn x)χ[−1,1] (x), then

 2 − 3|s|, |s| ≤ 1;
(f ∗ f )(x) = −2 + |s|, 1 < |s| ≤ 2;

0, otherwise.

and ∥f ∗ f ∥ = 8
3 while ∥f ∥ = 2, so ∥f ∗ f ∥ ̸= ∥f ∥2 .

4.3 Consider the algebra of analytic funions A0 (D). Show that conjugation f 7→ f is an involution on A0 (D) and
under which A0 (D) is a C ∗ -algebra. Show the map ∗ : f 7→ f ∗ (z) = f (z̄) is an involution, too. Is A0 (D) a
C ∗ -algebra with this involution?

Proof. e statement crossed out does not hold, as f is not necessarily analytic when f is analytic.
It is obvious that ∗ is an involution. Let f (z) = iz + 1, then f ∗ (z) = −iz + 1, so ∥f ∗ f ∥ = sup|z|≤1 |1 + z 2 | = 3
while ∥f ∥2 = 4. Hence A0 (D) with ∗ as involution is not a C ∗ -algebra.

8
4.4 Let A be a Banach algebra with involution ∗ and S be a subset of A. We say that S is regular if
(1) S is commutative, i.e., ab = ba for any a, b ∈ S;
(2) S is closed under involution, i.e., whenever a ∈ S it holds that a∗ ∈ S.
Obviously for any a ∈ S we have aa∗ = a∗ a. A regular subset is said to be maximal if it is not a proper subset of
any normal subset. Let B be a maximal regular subset of A , show that
(1) B is a closed commutative subalgebra of A ;
(2) ∀a ∈ B, it holds that σB (a) = σA (a).

Proof. (1) Let a, b ∈ B. It is clear that a+b commutes with B because each of a and b commutes with B. Hence
B is closed under addition since B is maximal. Similarly it can be shown that S is closed under multiplication
and scalar multiplication. e laws of associativity and distributivity are inherited from A . erefore B is a
commutative subalgebra.
Now we shall show that B is closed. Suppose an → a and {an } ⊆ B. From the continuity of multiplication,
we have that a commutes with B. Since B is maximal, it must hold that a ∈ B. erefore B is closed.
(2) Clearly σA (a) ⊆ σB (a). If λ ̸∈ σA a then (λe − a)−1 exists. Since (λe − a)−1 commutes with B and B is
maximal, we have (λe − a)−1 ∈ B and therefore λ ̸∈ σB (a). Hence σB (a) = σA (a).

4.5 Let A be a C ∗ -algebra, show that

(1) Let a be an hermitian element, then σ(a) ⊂ R1 ;
(2) If a is normal (aa∗ = a∗ a) then ∥a∥ = r(a);
(3) ∥a∥2 = r(aa∗ ).

Proof. Let B be the closure of the subalgebra generated by e, a, elements of form (λe − a)−1 for λ ∈ ρ(a). en
B is commutative and σB (a) = σA (a). We consider Gelfand transform on B for part (1) and (2).
(1) It follows immediately from Arens' Lemma.
(2) In fa, normality of a means that a and a∗ is commutative. Hence the proof of eorem 5.4.8 (3) holds, and
∥a∥ = ∥Γa∥ = r(a).
(3) It is clear that aa∗ is normal, hence by part (2), r(aa∗ ) = ∥aa∗ ∥ = ∥a∗ ∥2 = ∥a∥2 .

4.6 Let A be a C ∗ -algebra and a ∈ A . We say a is positive, denoted by a ≥ 0, if a is hermitian and σ(a) ⊆ [0, +∞].
Show that
(1) ∀a ∈ A , aa∗ ≥ 0;
(2) If a, b ∈ A , a ≥ 0, b ≥ 0 then a + b ≥ 0;
(3) For all a ∈ A , e + aa∗ is invertible in A .

Proof. (1) It is trivial that aa∗ is hermitian. As in the previous problem, we consider Gelfand transform in C ∗ (a),
which is commutative since a = a∗ . us Γaa∗ = ΓaΓa∗ = |Γa|2 ≥ 0, therefore aa∗ ≥ 0.
(2) First consider C ∗ (a), which is commutative. We claim that ∥λe−a∥ ≤ λ for all λ ≥ ∥a∥. In fa, Γ(λe−a) =
λ − Γ(a). Since a ≥ 0, Γ(a)(J) ∈ [0, λ], we have that Γ(λe − a)(J) ∈ [0, λ] and thus ∥λe − a∥ =
∥Γ(λe − a)∥ ≤ λ. Similarly for b ≥ 0, we can find µ such that ∥µe − b∥ ≤ µ. Hence ∥(λ + µ)e − (a + b)∥ ≤
∥λe − a∥ + ∥µe − b∥ ≤ λ + µ. Now in C ∗ (a + b),
λ + µ ≥ ∥λ + µ)e − (a + b)∥ = ∥Γ((λ + µ)e − (a + b))∥ ≥ |(λ + µ) − Γ(a + b)(J)| ≥ (λ + µ) − Γ(a + b)(J)
for all J, whence it follows that Γ(a + b)(J) ≥ 0 for all J, that is, a + b ≥ 0.
Similarly, by choosing an appropriate subalgebra, we may, without loss of generality, assume that A is com-
mutative. en Γ(a + b) = Γa + Γb ≥ 0. Hence a + b ≥ 0.

9
 (3) is is a dire corollary of part (1).

4.7 Let X be a Hilbert space and A a C ∗ -algebra of L(X ). Define

A c = {T ∈ L(X ) : T A = AT, ∀A ∈ A },

which is called the centre of A . Show that A c is a C ∗ -algebra and closed under weak topology.

Proof. It is easy to see that A c is closed under addition, multiplication, scalar multiplication and involution, whence
it follows that A c is a C ∗ -algebra. Now we shall show it is closed under weak topology. Suppose that Tn ⇀ T ,
{Tn } ⊆ A c . Note that X is a Hilbert space and thus reflexive, we have Tn x → T x for all x ∈ H. Hence
ATn x → AT x, that is, Tn Ax → AT x for all A ∈ A and x ∈ H. Recall that multiplication is continuous,
Tn A → T A, hence T Ax = AT x for all x ∈ H, that is exaly T A = AT for all A ∈ A . erefore T ∈ A c and
A c is closed under weak topology.

5 Normal Operators in Hilbert spaces

5.1 Let N be a normal operator in a Hilbert space. Show that
(1) If ϕ ∈ C(σ(N )) then σ(ϕ(N )) = ϕ(σ(N ));
(2) If ϕ ∈ C(σ(N )), ψ ∈ C(σ(ϕ(N ))), then (ψ ◦ ϕ)(N ) = ψ(ϕ(N )).

Proof. (1) Since λI − ϕ(N ) = (λ − ϕ)(N ), we have that λI − ϕ(N ) is invertible ⇔ ϕ(z) ̸= λ for all z ∈ σ(N )
⇔ λ ̸∈ ϕ(σ(N )).
(2) By part (1), ψ(ϕ(N )) is well-defined. It is clear that the conclusion holds when ψ is a polynomial of z and z̄.
For a general ψ, pick a sequence of polynomials ψn → ψ, then ψn ◦ ϕ → ψ ◦ ϕ, the conclusion follows from
the continuity of the isomorphism between AN and C(σ(N )).

5.2 Show that N is normal iff ∥N x∥ = ∥N ∗ x∥ for all x.

Proof. N ∗ N = N N ∗ ⇔ ⟨N ∗ N x, x⟩ = ⟨N N ∗ x, x⟩ ⇔ ⟨N x, N x⟩ = ⟨N ∗ x, N ∗ x⟩ ⇔ ∥N x∥2 = ∥N ∗ x∥2 .

5.3 Let N be a normal operator. Show that

(1) ∥N ∥ = sup{|λ| : λ ∈ σ(N )}, and if P is a polynomial then

∥P (N )∥ = sup{|P (λ)| : λ ∈ σ(N )}

(2) for A ∈ L(X ) it holds that ∥A∥2 = r(AA∗ ).

Proof. (1) When N is normal, P (N ) is normal, too. e conclusion follows from Problem 5.4.5(2).
(2) Problem 5.4.5(3).

5.4 Show that the produ of two positive operators is positive.

Proof. Notice that Γ(ab) = ΓaΓb.

5.5 Let A, B ∈ L(X ), 0 ≤ A ≤ B. Suppose that A and B are commutative, then A2 ≤ B 2 . However, this is not
necessarily true when A and B are not commutative.

10
 Proof. Note that B 2 − A2 = (B + A)(B − A) because A and B are commutative. e conclusion then follows
from the previous problem and Problem 5.4.6.
( ) ( )
1 2 1 4
Take X = R . Let A =
2
and B = , then B ≥ A ≥ 0 but AB ̸= BA. en B 2 − A2 =
( ) 2 5 2 10
4 32
, which is not positive.
10 79

5.6 Let N be a normal operator, then there exists P, Q ∈ L(X ), P is positive and unique, Q is unitary, such that

N = P Q = QP.

is is called polar decomposition of N .

Proof. Put p(z) = |z| and q(z) = z/|z| if z ̸= 0, q(0) = 1. en p and q are continuous funions on σ(N ).
Put P = Γ̃−1 p and Q = Γ̃−1 q. Since p ≥ 0, we know that P ≥ 0. Since q q̄ = 1, QQ∗ = Q∗ Q = I. Since
z = p(z)q(z), the relation N = P Q = QP follows from the symbolic calculus.
Now we prove that uniqueness of P . If N = QP , P positive and Q unitary, then N ∗ N = P ∗ Q∗ QP = P ∗ P = P 2 .
e uniquesness of P follows from the uniqueness of (N ∗ N ) 2 .
1

5.7 Let X be a locally compa topological space and H a Hilbert space. Definition 5.5.13 gives a speral family
(X , B, E), show that if ∆1 , ∆2 ∈ B then

E(∆1 ∩ ∆2 ) = E(∆1 )E(∆2 ).

Proof. First we show that if ∆1 ∩ ∆2 = then E(∆1 )E(∆2 ) = 0. Note that E(∆1 ∪ ∆2 ) = E(∆1 ) + E(∆2 ) is a
projeor, hence (E(∆1 ) + E(∆2 ))2 = E(∆1 ) + E(∆2 ), whence it follows that E(∆1 )E(∆2 ) = 0.
In the general case, E(∆1 ) = E(∆1 \ ∆2 ) + E(∆1 ∩ ∆2 ) and E(∆2 ) = E(∆1 \ ∆1 ) + E(∆1 ∩ ∆2 ). Hence

E(∆1 )E(∆2 ) = (E(∆1 \ ∆2 ) + E(∆1 ∩ ∆2 ))(E(∆1 \ ∆1 ) + E(∆1 ∩ ∆2 )).

Notice that ∆1 \ ∆2 , ∆2 \ ∆1 and ∆1 ∩ ∆2 are mutually disjoint, the conclusion follows from the expansion of the
right-hand side.

5.8 Let N be a normal operator and E is the associated speral family. en for any Borel set ∆ ⊆ C, E(∆) is
contained in the weak closure of the C ∗ algebra generated by N and N ∗ . Let S ∈ L(X ), SN = N S, show that
SE(∆) = E(∆)S.
L  e unit ball of a Banach space X is weak*-dense in the unit ball of X ∗∗ .
P  L  Suppose there exists x∗∗ ∈ X ∗∗ , ∥x∥X ∗∗ ≤ 1, x is not in the weak*-closure of B(X) ⊆ X ∗∗ .
Hence there exists a funional f ∈ X ∗ and a real number c, by Hahn-Banach eorem, such that ℜ⟨x, f ⟩ < c <
ℜ⟨x∗∗ , f ⟩ for all x in the unit ball of X. Since 0 is contained in the unit ball of X, c > 0. We can then divide by
c and replace f by c−1 f , and assume that there exists f ∈ X ∗ such that ℜ⟨x, f ⟩ < 1 < ℜ⟨x∗∗ , f ⟩. Since −ix is in
the unit ball of X whenever x is in it, this implies that ℑ⟨x, f ⟩ < 1 and thus |⟨x, f ⟩| ≤ 1 for all ∥x∥ ≤ 1. Hence
∥f ∥X ∗ ≤ 1, then 1 < |⟨x∗∗ , f ⟩| ≤ ∥x∗∗ ∥ ∥f ∥ ≤ 1. Contradiion.
L  Let X be a compa space and ∫ ϕ ∈ B(X).∫ en there exists a sequence of continuous funions {un } on
X such that ∥un ∥ ≤ ∥ϕ∥ for all n and un dm → ϕ dm for all m ∈ M (X).
P  L  Note that C(X) is a Banach space. e previous lemma tells us that the unit ball of C(X)
is weak*-dense in the unit ball of C(X)∗∗ = M (X)∗ . Identify B(X) with a subspace of M (X)∗ , and we are
done.

11
 Proof. By Lemma 2 and eorem 5.5.14, there exist {fn } ⊆ C(σ(N )) such that
∫ ∫
(fn (N )x, y) = fn (z) d(E(z)x, y) → χ∆ (z) d(E(z)x, y) = (χ∆ (N )x, y)

for all x and y. Since fn (N ) ∈ AN , the result above shows that χ∆ is in the weak closure of AN . Finally note
that χ∆ (N ) = E(∆ ∩ σ(N )) = E(∆). Also we have ∥fn ∥ ≤ ∥χ∆ ∥ = 1, thus by eorem 5.5.12(4), SE(∆) =
E(∆)S.

5.9 Let N be a normal operator. Prove that

(1) N is unitary ⇔ σ(N ) ⊂ S 1 ;
(2) N is self-adjoint ⇔ σ(N ) ⊂ R1 ;
(3) N is positive ⇔ σ(N ) ⊂ R1+ ;

Proof. (1) N is unitary ⇒ 1 = Γ(N N ∗ ) = (ΓN )(ΓN ∗ ) = ΓN ΓN = |ΓN |2 ⇒ σ(N ) ⊆ S 1 . Reversing the
procedure above, σ(N ) ⊆ S 1 ⇒ Γ(N N ∗ ) = 1 and similarly σ(N ) ⊆ S 1 ⇒ Γ(N ∗ N ) = 1. It follows from
the injeivity of Γ that N N ∗ = N ∗ N = I, i.e., N is unitary.
(2) N is self-adjoint ⇔ ΓN = ΓN ∗ = ΓN ⇔ ΓN ∈ R ⇔ σ(N ) ⊆ R1 .
(3) See eorem 5.5.5.

5.10 Suppose that N is a normal operator and σ(N ) is countable. en there is an orthonormal basis B = {y} ⊆ X ,
where y's are eigenveors of N , and Fourier expansion
∑
x= (x, y)y, ∀x ∈ X ,
y∈B

where the Fourier coefficients (x, y) = 0 except for countably many ones.

Proof. We claim that if x and y belong to different eigenvalues, say λ and µ, then ⟨x, y⟩ = 0. In fa, E({λ})X =
ker(λI − N ) and E({λ})X is orthogonal to E({µ})X .
Since ker(λI − T ) is closed, so we can choose an orthonormal basis. Combining those bases of each λ ∈ σ(N ),
we obtain an orthonormal set in H . We shall show that it is complete, i.e., x ⊥ {y}, or, x ⊥ ker(λI − T ) for all
λ implies that x = 0.
Let Pλ = E({λ}). Suppose that x ⊥ ker(λI − T ) = im Pλ , then x ∈ im Pλ⊥ = ker Pλ , that is, Pλ x = 0. Suppose
that σ(N ) = {λ1 , λ2 , . . . }, then

∑
n
x = Ix = E(σ(N )x = lim Pλk x = 0
n→∞
k=1

as desired. Fourier expansion follows from eorem 1.6.23 and 1.6.25.

5.11 Let N be a normal operator on H . Show that N is compa if and only if all following three conditions hold:
(1) σ(N ) is countable;
(2) If σ(N ) has a limit point, it must be 0;
(3) If λ ∈ σ(N ), λ ̸= 0, then dim E({λ})H < +∞.

12
 Proof. `Only if ': is is rather straightforward. Since N is compa, we have that σ(N ) \ {0} = σp (N ) \ {0} and
σp (N ) has at most one limit point 0 (eorem 4.3.1). Hence (1) and (2) hold, while (3) is just Fredholm eorem
(eorem 4.2.10(3)), i.e., dim ker(λI − N ) < +∞.
`If ': Suppose that |λ1 | ≥ |λ2 | ≥ ·∑ · · . Denote Pn = E({λ1 , . . . , λn }) and P0 be the projeor along ker N . Let
Nn = N Pn . en dim Nn H ≤ E(λi )H < +∞, which implies that Nn is finite-rank and thus compa. If
σ(N ) is finite then Nn = N for some n and N is therefore compa. Now assume that σ(N ) is infinite and thus
λn → 0. We want to show that Nn → N .
∑
Constru an orthonormal basis as in the previous problem, we have that x =
∑ (x, ei )ei for all x and N x =
λi (x, ei )ei . Note that Nn x is just a partial sum of N x, containing all terms up to λn (inclusive). en the
argument in Remark 1 after eorem 4.4.7 is valid, and we see that Nn → N and N is therefore compa.

5.12 Let N be a compa normal operator, show that

(1) ere exists λ, an eigenvalue of N , such that ∥N ∥ = |λ|;

(2) If ϕ ∈ C(σ(N )) and ϕ(0) = 0, then ϕ(N ) is compa.

Proof. (1) Suppose that |λ1 | ≥ |λ2 | ≥ ∑· · · . We claim that ∥N ∥ = |λ1 |. Problem 5.5.10 and Problem 5.5.11 tells
us that N is diagonalisable, N x = λi (x, ei )ei . Hence
∑ (∑ ) 21 (∑ ) 12

∥N x∥ = λi (x, ei )ei = |λi |2 |(x, ei )|2 ≤ |λ1 | |(x, ei )|2 ≤ |λ1 | ∥x∥,

hence ∥N ∥ ≤ |λ1 |. For x ∈ ker(λ1 I − N ), it holds that ∥N x∥ = ∥λ1 ∥, ∥x∥. erefore ∥N ∥ = |λ1 |.
(2) It is clear that ϕ(N ) is normal. en we shall verify the three conditions in the previous problem are satisfied.
Firstly, σ(ϕ(N )) = ϕ(σ(N )) is countable because σ(N ) is countable. Secondly, suppose y0 is a limit point of
σ(ϕ(N )) = ϕ(σ(N )), then there exists {xn } ⊆ σ(N ) (xn mutually different) such that ϕ(xn ) → y0 . Since
σ(N ) is compa, we can find a subsequence of xn , still denoted by xn , such that xn → x0 for some x0 ∈ σ(N ).
We know that x0 = 0, and thus y0 = ϕ(x0 ) = ϕ(0) = 0. irdly, note that Eϕ(N ) (λ) = EN (ϕ−1 (λ)) and
ϕ−1 (λ) is finite when λ ̸= 0. It follows immediately that dim Eϕ(N ) (λ)H < ∞ when λ ̸= 0.

5.13 Let N be a normal operator and E the speral family corresponding to N . Let ϕ ∈ C(σ(N )) and ω = ker ϕ.
Show that
ker ϕ(N ) = im E(ω).

Proof. ker ϕ(N ) = im Eϕ(N ) ({0}) = im EN (ϕ−1 ({0})) = im EN (ω), as it is not difficult to show that Eϕ(N ) (Ω) =
EN (ϕ−1 (Ω)) using Problem 5.5.1.

5.14 Let N be a normal operator, O an open set containing σ(N ) with a Jordan boundary. Suppose that ϕ is analytic on
a neighbourhood of σ(N ) and O is contained in the analytic domain of ϕ. Show that
∫
1
ϕ(N ) = ϕ(z)(zI − N )−1 dz.
2πi ∂O

Proof. It is known in eorem 2.6.9 that (zI−N )∫−1 is analytic on its domain. Next, we show that if F is an operator-
valued analytic funion over a domain Ω then ∫ ∂Ω F dz =∫ 0. In fa, let ϕ be any continuous funional, ϕ ◦ F
is analytic. It follows from continuity that ϕ( ∂Ω F dz) = ∂Ω ϕ ◦ F dz, which equals to 0 by Cauchy's ∫ eorem.
Note that it holds for any continuous funional ϕ, by Hahn-Banach eorem, it must hold that ∂Ω F dz = 0.
Hence for polynomial ϕ, we can replace ∂O by a possibly larger∑ circle outside |z| ≤ ∥N ∥. It is therefore easy to
∞
verify the desired equation, using expansion (zI − N )−1 = z −1 i=0 z −i Ai . Having established the equation for
polynomial ϕ, we can approximate a general ϕ by polynomials, completing the proof.

13
5.15 Let N be a normal operator and C a conneed component of σ(N ). Suppose that Γ ⊆ ρ(N ) is a Jordan curve, Γ
encloses C and contains no other sperum inside itself besides C. Show that
∮
1
E(C) = (zI − N )−1 dz.
2πi Γ
∮
Proof. Let P = 2πi1
Γ
(zI − N )−1 dz. First we shall show that P 2 = P∮. Choose Γ1 inside Γ∮such that Γ1 encloses
C also. Similar to the argument in the previous problem, we have that Γ1 (zI − N )−1 dz = Γ (zI − N )−1 dz.
∮ ∮
1 −1 1
2
P = (zI − N ) dz (wI − N )−1 dw
2πi Γ 2πi Γ1
∮ ∫
1
=− 2 (zI − N )−1 (wI − N )−1 dwdz
4π Γ Γ1
∮ ∫
1 (zI − N )−1 − (wI − N )−1
=− 2 dwdz
4π Γ Γ1 w−z
∮ ∫ ∮ ∮
1 −1 dw 1 −1 dz
= − 2 (zI − N ) dz − 2 (wI − N ) dw
4π Γ Γ1 w − z 4π Γ1 Γ z−w
∫
1
=0+ 2 2πi(wI − N )−1 dw
4π Γ1
∮
1
= (wI − N )−1 dw = P.
2πi Γ1

It is clear from the definition that P commutes with any bounded operator that commutes with N . Now we
shall show that the sperum of N restried to im P is contained in C, that is, λI − N is invertible on im P for
λ ∈ σ(N ) \ C. Consider ∮
1 1
Q= (zI − N )−1 dz
2πi Γ λ − z
It is clear that QN = N Q and thus QP = P Q. In fa,
∮
1 (λ − z)I + (zI − N )
Q(λI − N ) = (λI − N )Q = (zI − N )−1 dz
2πi Γ λ−z
∮ ∮
1 I dz
= (zI − N )−1 dz +
2πi Γ 2πi Γ λ − z
= P + 0 = P.

Note that P is the identity map on im P , hence Q is the the inverse of (λI − N ) on im P . Also, the sperum of
N restried to ker P is contained in σ(N ) \ C, which can be shown similarly by replacing C by σ(N ) \ C (the
conneedness of C is not essential, separation of two compa sets is).
Now we look at E(C), which satisfies all the properties that we have proved for P . As a consequence, it is easy to
prove that P E(C) = P and E(C)P = E(C). erefore P = E(C).

6 Applications to Singular Integral Operators

No Exercises.

14