Banach Algebras and Spectral Theory for

Operators on a Banach Space

CRISTIAN MARIO GALERA MARTÍNEZ

ÁLGEBRA DE BANACH

UNIVERSIDAD DE ALMERÍA
 Contents
Chapter 1. Banach Algebras and Spectral Theory for Operators on a Banach
Space 4
1.1. Elementary Properties and Examples 4
1.2. Ideals and Quotients 5
1.3. The Spectrum 5
1.4. The Riesz Functional Calculus 6
1.5. Dependence of the Spectrum on the Algebra 10
1.6. The Spectrum of a Linear Operator 11
1.7. The Spectral Theory of a Compact Operator 12
1.8. Abelian Banach Algebras 13
1.9. The Group Algebra of a Locally Compact Abelian Group 15
Bibliography 17

3
 CHAPTER 1

Banach Algebras and Spectral Theory for

Operators on a Banach Space
1.1. Elementary Properties and Examples

An algebra over F is a vector space A over F that also has a multiplication

dened on it that makes A into a ring such that if α ∈ F and a, b ∈ A, α(ab) =
(αa)b = a(αb).

Definition 1. A Banach algebra is an algebra A over F that has a norm ||.||

relative to which A is a Banach space and such that for all a, b ∈ A,

||ab|| ≤ ||a||||b|| (1)

.
If A has an identity, e, then it is assumed that ||e|| = 1.
The fact that (1) is satised is not essential. If A is an algebra and has a norm
relative to which A is a Banach space and is such that the map of A × A → A
dened by (a, b) 7→ ab is continuous, then there is an equivalent norm on A satises
(1).
If A has an identity e, then the map α 7→ αe is an isomorphism of F into A
and ||αe|| = |α|. So it will be assumed that F ⊆ A via this identication. Thus the
identity will be denoted by 1.
Proposition 2. If A is a Banach algebra without an identity, let A1 = A × F.
Dene algebraic operations on A1 by:
(i) (a, α) + (b, β) = (a + b, α + β);
(ii) β(a, α) = (βa, βα);
(iii) (a, α)(b, β) = (ab + αb + βa, αβ).
Dene ||(a, α)|| = ||a|| + |α|. Then A1 with this norm and the algebraic op-
erations dened in (i), (ii) and (iii) is a Banach algebra with identity (0, 1) and
a 7→ (a, 0) is a isometric isomorphism of A into A1 .
Proposition 3. If {Ai } is a collection of Banach algebras, ⊕0 Ai and ⊕∞ Ai
are Banach algebras.

4
 1.3. THE SPECTRUM 5

1.2. Ideals and Quotients

The next proof is based on the geometric series.

Lemma 4. If A is a Banach algebra with identity and x ∈ A such that ||x−1|| <
1, then x is invertible.
Theorem 5. If A is a Banach algebra with identity, Gl = {a ∈ A : a is lef t invertible},
Gr = {a ∈ A : a is right invertible}, and G = {a ∈ A : a is invertible}, then Gl ,
Gr and G are open subsets of A. Also, the map a 7→ a−1 of G → G is continuous.
Proposition 6. If A is a Banach algebra with identity, then every proper left,
right, or bilateral ideal is contained in a maximal ideal of the same type.
Theorem 7. If A is a Banach algebra and M is a proper closed ideal in A,
the A/M is a Banach algebra. If A has an identity, so does A/M.

It may be that A/M has an identity even if A does not. For example, let
A = C0 (R) and let M = {Φ ∈ C0 (R) : Φ(x) = 0 when |x| ≤ 1}. If Φ0 ∈ C0 (R)
such that Φ0 (x) = 1 for |x| ≤ 1, then Φ0 + M is and identity for A/M. In fact, if
Φ ∈ C0 (R), (ΦΦ0 − Φ)(x) = 0 if |x| ≤ 1. Hence (Φ + M)(Φ0 + M) = Φ + M.

1.3. The Spectrum

Definition 8. If A is a Banach algebra with identity and a ∈ A, the spectrum

of a, denoted by σ(a), is dened by:

σ(a) = {α ∈ F : a − α is not invertible}

.
The left spectrum, σl (a), is the set {σ ∈ F : a − α is not lef t invertible}; the
right spectrum, σr (a), is dened similarly.
The resolvent set of a is dened by ρ(a) = F \ σ(a). The left and right resolvent
of a are ρl (a) = F \ σl (a) and ρr (a) = F \ σr (a).
Theorem 9. If A is a Banach algebra over C with an identity, then for each
a ∈ A, σ(a) is a nonempty compact subset of C. Moreover, if |α| > ||a||, α ∈
/ σ(a)
and z 7→ (z − a)−1 is an A-valued analytic function dened on ρ(a).

Because the spectrum of an element of a complex Banach algebra is not empty,

the following claim is made.
Claim 10. Henceforward, all Banach spaces and all Banach algebras are over
C.
Definition 11. If A is a Banach algebra with identity and a ∈ A, the spectral
radius of a, r(a), is dened by:

r(a) = sup{|α| : α ∈ σ(a)}.

Because σ(a) ≤ 0 and is bounded, r(a) is well dened and nite; because σ(a)
is compact, this sumpremum is attained.
 1.4. THE RIESZ FUNCTIONAL CALCULUS 6

 
00
Let A = M2 (C) and let A = . Then A2 = 0 and σ(A) = 0; so r(A) = 0.
10
So it is possible to have r(A) = 0 with A ≤ 0.
Proposition 12. If A is a Banach algebra with identity and a ∈ A, lim||an ||1/n
exists and:

r(a) = lim||an ||1/n

.
Proposition 13. Let A be a Banach algebra with identity and let a ∈ A.
(a) If α ∈ ρ(a), then dist(α, σ(a)) ≤ ||(α − a)−1 ||−1
(b) If α, β ∈ ρ(a), then

(α − a)−1 − (β − a)−1 = (β − α)(α − a)−1 (β − a)−1

= (β − α)(β − α)−1 (α − a)−1 .

The identity in part (b) of the proceeding proposition is called the resolvent
identity and the function α 7→ (α − a)−1 of ρ(a) → A is called the resolvent of a.

1.4. The Riesz Functional Calculus

If γ is a closed recticable curve in C and a ∈

/ {γ} ≡ {γ(t) : 0 ≤ t ≤ 1}, then
the winding number of γ about a is dened to be the number:
Z
1 1
n(γ; a) = dz
2πi γ z−a
.
The number n(γ; a) is always an integer and is constant on each component of
C \ {γ} and vanishes on the unbounded component of C \ {γ}.
If γ is a rectiable curve in G and Rf is a continuous function dened in a
neighborhood of γ with values in H, then γ f can be dened as for a scalar-valued
f as the limit in H of sums of the form:
X
[γ(tj ) − γ(tj−1 )]f (γ(tj ))
j
,
where {t0 , t1 , ..., tn } is a partition of [0, 1]. Hence γ f = 0 f (γ(t))dγ(t) ∈ H.
R R1

It is easy to see that for every x∗ ∈ H∗ , h γ f, x∗ i = γ hf (·), x∗ i.

R R

Theorem 14. Cauchy´s Theorem:

If H is a Banach space, G is an open subset of C, f : G → HPis an analytic
function, and γ1 , ..., γm arePclosed rectiable curves in G such that m
j=1 n(γj ; a) =
0 for all a in C \ G, then j=1 γj f = 0.
m R
 1.4. THE RIESZ FUNCTIONAL CALCULUS 7

If x∗ ∈ H∗ , then h mj=1 γj f, x i = Pj=1 Rγj hf (·), x i = 0 by the

∗ m ∗
P R P R
Proof.

scalar-valued version of Cauchy´s Theorem. Hence j=1 γj f = 0.

m

Theorem 15. Cauchy´s Integral Formula:
If H is a Banach space, G is an open subset of C, f : G → H is analytic, γ
is a closed rectiable curve in G such that n(γ; a) = 0 for every a ∈ C \ G, and
λ ∈ G \ {γ}, then for every integer k ≥ 0,
Z
k!
n(γ, λ)f (k)
(λ) = (z − λ)−(k+1) f (z)dz.
2πi γ

Definition 16. A closed rectiable curve γ is positively oriented if for every

a ∈ G \ {γ}, n(γ; a) is either 0 or 1. In this case the inside of γ , denoted by ins γ ,
is dened by:

ins γ ≡ {a ∈ C \ {γ} : n(γ; a) = 1}

.
The outside of γ , denoted by out γ , is dened by:

out γ ≡ {a ∈ C \ {γ} : n(γ; a) = 0}

Thus C = {γ} ∪ insγ ∪ outγ .
A curve γ : [0, 1] → C is simple if γ(s) = γ(t) implies that either s = t or s = 0
and t = 1.
If Γ = {γ1 , ..., γm } is a collection of closed rectiable curves,
S then Γ is positively
oriented if: (a) {γi } ∩ {γj } = ∅ for i 6= j ; (b) for a ∈ C \ m j=1 {γj }, n(Γ; a) ≡
j=1 n(γj ; a) is either 0 or 1; (c) each γj is a simple curve. The inside of Γ, ins Γ,
Pm
is dened by:

ins Γ ≡ {a : n(Γ; a) = 1}
.
The outside of Γ, out Γ, is dened by:

out Γ ≡ {a : n(Γ; a) = 0}
.
Let {Γ} ≡ ∪{γj : 1 ≤ j ≤ m}.

Proposition 17. If G is an open subset of C and K is a compact subset of G,

then there is a positively oriented system of curves Γ = {γ1 , ..., γm } in G \ K such
that K ⊆ ins Γ and C\ ⊆ out Γ. The curves γ1 , ..., γm can be found such that they
are innitely dierentiable.
Proposition 18. Let A be a Banach algebra with identity, let a ∈ A, and let G
be an open subset of C such that σ(a) ⊆ G. If Γ = {γ1 , ..., γm } and Λ = {λ1 , ..., λk }
are two positively oriented collections of curves in G such that σ(a) ⊆ ins Γ and
σ(a) ⊆ ins Λ and if f : G → C is analytic, then
Z Z
f (z)(z − a)−1 dz = f (z)(z − a)−1 dz
. Γ Λ
 1.4. THE RIESZ FUNCTIONAL CALCULUS 8

Hol(a) = all of the functions that are analytic in a neighborhood of σ(a). Note
that Hol(a) is an algebra where if, f, g ∈ Hol(a) and f and g have domains D(f )
and D(g), then f g and f + g have domain D(f ) ∩ D(g). Hol(a) is not, however, a
Banach algebra.
Theorem 19. The Riesz Functional Calculus:
Let A be a Banach algebra with identity and let a ∈ A.
(a) The map fP7→ f (a) of Hol(a) → A is an algebra homomorphism.
(b) If f P
(z) = k=0 αk z k has radius of convergence > r(a), then f ∈ Hol(a)
∞

and f (a) = k=0 αk ak .

∞

(c) If f (z) ≡ 1, then f (a) = 1.

(d) If f (z) = z for all z , f (a) = a.
(e) If f, f1 , f2 , ... are all analytic on G, σ(a) ⊆ G, and fn (z) → f (z) uniformly
on compact subsets of G, then ||fn (a) − f (a)|| → 0 as n → ∞.
Proof. (a) Let f, g ∈ Hol(a) and let G be an open neighborhood of σ(a) on
witch both f and g are analytic. Let Γ be a positively oriented system of closed
curves G such that σ(a) ⊆ Γ. Let Λ be a positively system of closed curves in G
such that (ins Γ) ∪ {Γ} = cl(ins Γ) ⊆ ins Λ. Then
Z  Z 
1 −1 −1
f (a)g(a) = f (z)(z − a) dz g(ζ)(ζ − a) dζ
4π 2 Γ Λ
Z Z
1
=− 2 f (z)g(ζ)(z − a)−1 (ζ − a)−1 dζdz
4π Γ Λ
by Proposition 13

(z − a)−1 − (ζ − a)−1
Z Z  
1
=− f (z)g(ζ) dζdz
4π 2 Γ Λ ζ −z
Z Z 
1 g(ζ)
=− 2 f (z) dz (z − a)−1 dz
4π Γ Λ ζ −z
Z Z 
1 f (z)
+ 2 g(ζ) dz (ζ − a)−1 dζ.
4π Λ Γ ζ −z
But for ζ on Λ, ζ ∈ out Γ and hence [f (z)/(ζ−z)]dz = 0 (Cauchy´s Theorem).
R
Γ
If z ∈ {Γ}, then z ∈ ins Λ and so Λ [g(ζ)/(ζ − z)]dζ = 2πig(z). Hence
R

Z
1
f (a)g(a) = f (z)g(z)(z − a)−1 dz = (f g)(a).
2πi Γ
The proof that (αf + βg)(a) = αf (a) + βg(a) is left to the reader.
(c) and (d). Let f (z) = z k , k ≥ 0. Let γ(t) = Rexp(2πit), 0 ≤ t ≤ 1, where
R > ||a||. So σ(a) ⊂ ins γ , and hence
Z
1
f (a) = z k (z − a)−1 dz
2πi γ
Z  −1
1 a
= z k−1 1 − dz
2πi γ z
 1.4. THE RIESZ FUNCTIONAL CALCULUS 9

Z ∞
1 X
= z k−1 an /z n dz,
2πi γ n=0
since ||a/z|| < 1 for |z| = R. Since this innite series converges uniformly for z
on γ ,
∞  Z 
X 1 1
f (a) = an .
2πi γ z n−k+1 dz
k=0

If n 6= k then z −(n−k+1) has a primitive and hence γ z −(n−k+1) dz = 0. For

R

n = k this integral becomes γ z −1 dz = 2πi. Hence f (a) = ak .

R

(e) Let Γ = {γ1 , ..., γm } be a positively oriented system of closed curves in G

such that σ(a) ⊆ Γ. Fix 1 ≤ k ≤ m; then
 Z Z 
−1 −1 
 

 fn (z)(z − a) dz − f (z)(z − a) 
γk γk
 Z 1 
−1
 
= 
  [fn (γk (t)) − f (γk (t))][γk (t) − a] dγk (t)
0
Z 1
≤ |fn (γk (t)) − f (γk (t))|||[γk (t) − a]−1 ||d|γk |(t).
0
Now t 7→ ||[γk (t) − a]−1 || is continuous on [0, 1] and hence bounded by some
constant, say M . Thus:
 Z Z 
fn (z)(z − a)−1 dz − f (z)(z − a)−1 
 


γk γk

≤ M ||γk ||max{|fn (z) − f (z)| : z ∈ {γk }},

where ||γk || is the total variation (length) of γk . By hypothesis it follows that
||fn (a) − f (a)|| →P 0 as n → ∞.
(b) If p(z)
Pn k=0kαk z is a polynomial,
=
n k
P then (a), (c) and (d) combine to ve
that p(a) = k=0 αk a . Now let f (z) = ∞ k=0 αk z k
have radius of convergence
R > r(a), the spectral radius of a. If pn (z) = k=0 αk z k , pn (z) → f (z) uniformly
Pn
on compact subsets of {z : |z| < R}. By (e), pn (a) → f (a). So (b) follows. 

Proposition 20. Let A be a Banach algebra with identity and let a ∈ A. Let
τ : Hol(a) → A be a homomorphism such that (a) τ (1) = 1, (b) τ (z) = a, (c)
if {fn } is a sequence of analytic functions on an open set G such that σ(a) ⊆ G
and fn (z) → f (z) uniformly on compact subsets of G, then τ (fn ) → τ (f ). Then
τ (f ) = f (a) for every f in Hol(a).
Proposition 21. If a, b ∈ A, ab = ba, and f ∈ Hol(a), then f (a)b = bf (a).
Theorem 22. The Spectral Mapping Theorem:
If a ∈ A and f ∈ Hol(a), then

σ(f (a)) = f (σ(a)).

 1.5. DEPENDENCE OF THE SPECTRUM ON THE ALGEBRA 10

Proof. If α ∈ σ(a), let g ∈ Hol(a) such that f (z) − f (α) = (z − α)g(z). If it

were the case that f (α) ∈ / σ(f (a)), then (a − α) would be invertible with inverse
g(a)[f (a) − f (α)]−1 . Hence f (α) ∈ σ(f (a)); that is, f (σ(a)) ⊆ σ(f (a)).
Conversely, if β ∈/ f (σ(a)), then g(z) = [f (z)−β]−1 ∈ Hol(a) and so g(a)[f (a)−
β] = 1. Thus β ∈ / σ(f (a)); that is, σ(f (a)) ⊆ f (σ(a)). 

This section closes with an application of the functional calculus that is typical.
Proposition 23. Suppose a ∈ A and σ(a) = F1 ∪ F2 , where F1 and F2 are
disjoint nonempty closed sets. Then there is a nontrivial idempotent e ∈ A such
that
(a) if ba = ab, then be = eb;
(b) if a1 = ae and a2 = a(1 − e), then a = a1 + a2 and a1 a2 = a2 a1 = 0;
(c) σ(a1 ) = F1 ∪ {0}, σ(a2 ) = F2 ∪ {0}.

1.5. Dependence of the Spectrum on the Algebra

If ∂D = {z ∈ C : |z| = 1}, let B = the uniform closure of the polynomials in

C(∂D). If A = C(∂D), then the spectrum of z as an element of A is ∂D. That is,

σA (z) = ∂D.
Now z ∈ B and so it has a spectrum as an element of this algebra; denote this
spectrum by σB (z). There is no reason to believe that σB (z) = σA (z). In fact, they
are not equal.

Example 24. If B = the closure in C(∂D) of the polynomials in z , then σB (z) =

clD.
Thus the spectrum not only depends on the element of the algebra, but also on
the algebra. To phrase these results it is necessary to introduce the polynomially
convex hull of a compact subset of C.

Definition 25. It A is a set and f : A → C, dene

||f ||A ≡ sup{|f (z)| : z ∈ A}

.
If K is a compact subset of C, dene the polynomially convex hull of K to be
the set K ∧ given by

K ∧ ≡ {z ∈ C : |p(z)| ≤ ||p||K f or every polynomial p}

.
The set K is polynomially convex if K = K ∧ .
Note that the polynomially convex hull of ∂D is clD.
 1.6. THE SPECTRUM OF A LINEAR OPERATOR 11

Proposition 26. If K is a compact subset of C, then C \ K ∧ is the unbounded

component of C \ K . Hence K is polynomially convex if and only if C \ K is
connected.
Theorem 27. If A and B are Banach algebras with a common identity such
that B ⊆ A and a ∈ B, then
(a) σA (a) ⊆ σB (a) and ∂σB (a) ⊆ ∂σA (a).
(b) σA (a)∧ = σB (a)∧ .
(c) If G is a hole of σA (a), then either G ⊆ σB (a) or G ∩ σB (a) = ∅.
(d) If B is the closure in A of all polynomials in a, then σB (a) = σA (a)∧ .

1.6. The Spectrum of a Linear Operator

Proposition 28. (a) If I is a Banach space and A ∈ B(I), σ(A∗ ) = σ(A).

(b) If H is a Hilbert space and A ∈ B(H), σ(A∗ ) = σ(A)∗ , where for any subset
∆ of C, ∆∗ ≡ {z̄ : z ∈ ∆}.
Definition 29. If A ∈ B(I), the point spectrum of A, σp (A), is dened by

σp (A) ≡ {λ ∈ C : ker(A − λ) 6= (0)}

.
Definition 30. If A ∈ B(I), the proximate point spectrum of A, σap (A), is
dened by

σap (A) ≡ {λ ∈ C : there is a sequence {xn } ∈ I such that ||xn || = 1 f or all n and ||(A−λ)xn || → 0}.

Note that σp (A) ⊆ σap (A).

Proposition 31. If A ∈ B(I) and λ ∈ C, the following statements are equiv-
alent.
(a) λ ∈/ σap (A).
(b) ker(A − λ) = (0) and ran(A − λ is closed.
(c) There is a constant c > 0 such that ||(A − λ)x|| ≥ c||x|| for all x.
Proposition 32. If 1 ≤ p ≤ ∞, dene S : lp → lp by S(x1 , x2 , ...) =
(0, x1 , x2 , ...). Then σ(S) = clD, σp (S) = ∅, and σap (S) = ∂D. Moreover, for
|λ| < 1, ran(S − λ) is closed and dim[lp /ran(S − λ)] = 1.
Proposition 33. If A ∈ B(I), then ∂σ(A) ⊆ σap (A).
Claim 34. ||(A − λn )−1 || → ∞ an n → ∞.
Lemma 35. If λ0 is an isolated point of σ(A), then
∞
X
(z − A)−1 = (z − λ0 )n An
n=−∞
for 0 < |z − λ0 | < r0 = dist(λ0 , σ(A) \ {λ}), where
 1.7. THE SPECTRAL THEORY OF A COMPACT OPERATOR 12

Z
1
An = (z − λ0 )−n−1 (z − A)−1 dz
2πi γ
for γ = any circle centered at λ0 with radius < r0 .
Proposition 36. If λ0 is an isolated point of σ(A), then λ0 is a pole of (z −
A)−1 of order n if and only if (λ0 − A)n E(λ0 ) = 0 and (λ0 − A)n−1 E(λ0 ) 6= 0.
Corollary 37. If λ0 is an isolated point of σ(A) and is a pole of (z − A) ,
−1

then λ0 ∈ σp (A).
Claim 38. |kn (x, y)| ≤ (||k||n∞ /((n − 1)!)(x − y)n−1 ) for y < x.

1.7. The Spectral Theory of a Compact Operator

Recall that for a Banach space I , B0 (I) is the algebra of all compact operators.
This Banach algebra has no identity, so if A ∈ B0 (I), then σ(A) refers to the
spectrum of A as an element of B(I).
Lemma 39. If A ∈ B0 (I), λ 6= 0, and ker(A − λ) = (0), then ran(A − λ) is
closed.
Lemma 40. If A ∈ B0 (I), λ 6= 0, and λ ∈ σ(A), then either λ ∈ σp (A) or
λ ∈ σp (A∗ ).
Lemma 41. If M ≤ N , M = 6 N , and  > 0, then there is a y in N such that
||y|| = 1 and dist(y, M) ≥ 1 − .

If M and N are nite dimensional in the proceeding lemma, then y can be

chosen in N such that ||y|| = 1 and dist(y, M) = 1.
Lemma 42. If A ∈ B0 (I) and {λn } is a sequence of distinct elements in σp (A),
then limλn = 0.
Theorem 43. Theorem of F.Riesz:
If dimI = ∞ and A ∈ B0 (I), then one and only one of the following possibilities
occurs.
(a) σ(A) = {0}.
(b) σ(A) = {0, λ1 , ..., λn }, where for 1 ≤ k ≤ n, λk 6= 0, each λk is an eigen-
value of A, and dim(ker(A − λk )) < ∞.
(c) σ(A) = {0, λ1 , λ2 , ...}, where for each k ≥ 1, λk is an eigenvalue of A,
dim(ker(A − λk )) < ∞, and, moreover, liimλk = 0.
Proof. To proof will use the lasts lemmas. The rst step is to establish the
following. 
Claim 44. If λ ∈ σ(A) and λ 6= 0, then λ is an isolated point of σ(A).

In fact, if {λn } ⊆ and λn → λ, then each λn belongs to either σp (A) or σp (A∗ )

(Lemma 40). So either there is a subsequence λnk that is contained in σp (A) or
there is a subsequence contained in σp (A∗ ). If {λnk } ⊆ σp (A), then Lemma 42
implies λnk → 0, a contradiction. If {λnk } ⊆ σp (A∗ ), then the fact that A∗ is
compact gives the same contradiction. Thus λ must be isolated if λ 6= 0.
 1.8. ABELIAN BANACH ALGEBRAS 13

Claim 45. If λ ∈ σ(A) and λ 6= 0, then λ ∈ σp (A) and dim(ker(A − λ)) < ∞.

By (Lemma 43), λ is an isolated point of σ(A) so that E(λ) can be dened as

in:
Z
1
E(∆) = E(∆; A) = (z − A)−1 dz,
2πi Γ
where Γ is a positively oriented Jordan system such that ∆ ⊆ insΓ and σ(A) \
∆ ⊆ outΓ, is an idempotent. Let Iλ = E(λ)I and Aλ = A|Iλ . Thus σ(Aλ ) = {λ}.
Thus Aλ is an invertible compact operator. Thus dim(Iλ ) < ∞. If n = dim(Iλ ),
then Aλ − λ is a nilpotent operator on an n-dimensional space. Thus (Aλ − λ)n = 0.
Let v = the positive integer such that (Aλ − λ)v = 0 but (Aλ − λ)v−1 6= 0. Let
x ∈ Iλ such that 0 6= (Aλ − λ)v−1 x = y ; then (A − λ)y = 0. Thus λ ∈ σp (A).
Also, ker(A − λ) ∈ LatA and A|ker(A − λ) is compact. But Ax = λx for all
x ∈ ker(A − λ), so dim(ker(A − λ)) < ∞.
Now for the dénouement. If dim(I) = ∞ and A ∈ B0 (I) then A cannot be
invertible. Thus 0 ∈ σ(A). If λ ∈ σ(A) and λ 6= 0, then Claim 45 says that
λ ∈ σp (A) and dim(ker(A − λ)) < ∞. So if σ(A) is nite, either (a) or (b) of
(Theorem 43: F.Riesz) hold. If σ(A) is innite, then Claim 44 implies that σ(A) is
countable. So let σ(A) = {0, λ1 , λ2 , ...}. By Lemma 42 and Claim 45, (c) holds.
Theorem 46. The Fredholm Alternative:
If A ∈ B0 (I), λ ∈ C and λ 6= 0, then ran(A−λ) is closed and dim(ker(A−λ)) =
dim(ker(A − λ)∗ ) < ∞.

Proof. It suces to assume that λ ∈ σ(A). Put ∆ = σ(A) \ {λ}, Iλ = E(λ)I ,

I∆ = E(∆)I , Aλ = A|Iλ and A∆ = A|I∆ . Now λ ∈ / ∆ = σ(A∆ ), so A∆ − λ is
invertible. Thus ran(A∆ − λ) = I∆ . Hence ran(A − λ) = (A − λ)Iλ + (A − λ)I∆ =
ran(Aλ − λ) + I∆ . Since dimIλ < ∞, ran(A − λ) is closed.
Also note that

I/ran(A − λ) = (I∆ + Iλ )/[ran(Aλ − λ) + I∆ ] ≈ Iλ /ran(Aλ − λ).

Since dim(I) < ∞, dim[I/ran(A − λ)] = dim(Iλ − dim(ran(Aλ − λ))) =
dim(ker(Aλ − λ)) = dim(ker(A − λ)) < ∞since ker(A − λ) ⊆ I . But [I/ran(A −
λ)]∗ = [ran(A−λ)]⊥ = ker(A−λ)∗ . Hence dim(ker(A−λ)) = dim(ker(A−λ)∗ . 

1.8. Abelian Banach Algebras

Recall that it is assumed that every Banach algebra is over C. Also assume
that all Banach algebras contain an identity.
Theorem 47. If A is a Banach algebra that is also a division ring, then A =
C (≡ {λ1 : λ ∈ C}).
Proof. If a ∈ A, then σ(a) 6= ∅. If λ ∈ σ(a), then a − λ has no inverse. But
A is a division ring, so a − λ = 0. That is, a = λ. 
 1.8. ABELIAN BANACH ALGEBRAS 14

Proposition 48. If A is an abelian Banach algebra and M is a maximal ideal,

then there is a homomorphism h : A → C such that M = kerh. Conversely, if h :
A → C is a nonzero homomorphism, then kerh is a maximal ideal. Moreover, this
correspondence h 7→ kerh between homomorphism and maximal ideals is bijective.
Corollary 49. If A is an abelian Banach algebra and h : A → C is a homo-
morphism, then h is continuous.
Proposition 50. If A is abelian and h : A → C is a non-zero homomorphism,
then ||h|| = 1.
Definition 51. If A is an abelian Banach algebra, let Σ = the collection of
all nonzero homomorphisms of A → C. Give Σ the relative weak* topology that it
has as a subset of A∗ . Σ with this topology is called the maximal ideal space of A.
Theorem 52. If A is an abelian Banach algebra, then its maximal ideal space
Σ is a compact Hausdor space. Moreover, if a ∈ A, then σ(a) = Σ(a) ≡ {h(a) :
h ∈ Σ}.
Example 53. Here is one that is a little more than an example. If X is compact
and x ∈ X , let δx : C(X) → C be dened by δx (f ) = f (x). It is easy to see that δx
is a homomorphism on the algebra C(X).
Theorem 54. If X is a compact and Σ is the maximal ideal space of C(X),
then the map x 7→ δx is a homeomorphism of X onto Σ.
It follows from the preceding theorem that the maximal ideals of C(X) are all
of the form {f ∈ C(X) : f (x) = 0} for some x ∈ X .
Definition 55. Let A be an abelian Banach algebra with maximal ideal space
Σ. If a ∈ A, then the Gelfrand transform of a is the function â : Σ → C dened by
â(h) = h(a).
Theorem 56. If A is an abelian Banach algebra with maximal ideal space Σ
and a ∈ A, then the Gelfrand transform of a, â, belongs to C(Σ). The map a 7→ â
of A into C(Σ) is a continuous homomorphism of A into C(Σ) of norm 1 and its
kernel is
\
{M : M is a maximal ideal of A}.
Moreover, for each a ∈ A,

||â||∞ = limn→∞ ||an ||1/n .

The homomorphism a 7→ â of A into C(Σ) is called the Gelfand transform of

A. The kernel of the Gelfand transform is called the radical of A, radA. So

radA = ∩{M : M is a maximal ideal of A}.

Recall that if τ : X → Y is a homeomorphism, then A : C(Y ) → C(X) dened
by Af = f ◦ τ is an isometric isomorphism. Denote the relationship between A and
τ by A = τ ] .
 1.9. THE GROUP ALGEBRA OF A LOCALLY COMPACT ABELIAN GROUP 15

Proposition 57. If A is an abelian Banach algebra with identity and a is a

generator of A, then here is a homeomorphism τ : Σ → σ(a) such that if γ : A →
C(Σ) is the Gelfand transform and p is polynomial, then γ(p(a)) = τ ] (p).

1.9. The Group Algebra of a Locally Compact Abelian Group

If G is a locally compact abelian group and m is Haar measure on G, then

L1 (G) ≡ L1 (m) is a Banach algebra, where for f, g ∈ L1 (G) the product f ∗ g is
the convolution of f and g :
Z
f ∗ g(x) = f (xy −1 g(y))dy
. G

Note that dy is used to designate integration with respect to R m rather than

dm(y). Because G is abelian, L1 (G) is abelian. In fact, g ∗f (x) = g(xy −1 f (y))dy .
If y −1 x is substituted for y in this integral, the value of the integral
R does not change
because Haar measure is translation invariant. Hence g ∗f (x) = g(y)f (y −1
x)dy =
g(y)f (xy −1 )dy = f ∗ g(x).
R

Let e denote the identity of G. If G is discrete, then δe ∈1 (G) and δe is an

identity for L1 (G). If G is not discrete, then L1 (G) does not have an identity.
Proposition 58. If G is a topological group and f : G → C is a continuous
function with compact support, then for any  > 0 there is a neighborhood U of e
such that |f (x) − f (y)| <  whenever x−1 y ∈ U .

One can rephrase Proposition 58 by saying that continuous functions on a

topological group that have compact support are uniformly continuous.
In the next result it is the case p = 1 which is of principal interest for us at
this time.

Proposition 59. If G is a locally compact group, 1 ≤ p < ∞, and f ∈ Lp (G),

then the map x 7→ fx is a continuous function from G into Lp (G).

The aim of this section is to discuss the homomorphism on L1 (G) when G is

abelian and to examine the Gelfand transform.
Proposition 60. If f ∈ L1 (G) and  > 0, then there is a neighborhood U of
eRsuch that if g is a non-negative Borel function on G that vanishes o U and has
g(x)dx = 1, then ||f − f ∗ g||1 < .

R Corollary 61. There is a net ei of non-negative functions in L1 (G) such that

ei dm = for all i and ||ei ∗ f − f ||1 → 0 for all f ∈ L1 (G).
Corollary 62. If h : L (G) → C is a nonzero homomorphism, then h is
1

bounded and ||h|| = 1.

Even though Haar measure on most of the popular examples is σ-nite, this is
not true in general.
 1.9. THE GROUP ALGEBRA OF A LOCALLY COMPACT ABELIAN GROUP 16

Theorem 63. If G is a locally compact abelian group and γ : G → T is a

continuous homomorphism, dene fˆ(γ) by
Z
fˆ(γ) = f (x)γ(x−1 )dx

for every f ∈ L1 (G). Then f 7→ fˆ(γ) is a nonzero homomorphism on L1 (G).

Conversely, if h : L1 (G) → C is a nonzero homomorphism, there is a continuous
homomorphism γ : G → T such that h(f ) = fˆ(γ).
Proposition 64. Γ is a locally compact abelian group.

Clearly Γ is an abelian group and we know that Γ is a locally compact space.

It must be shown that Γ is a topological group. To do this we rst see a lemma.

Lemma 65. (a) The map (x, y) 7→ γ(x) of G × Γ → T is continuous.

(b) If {γi } is a net in Γ and γi → γ in Γ, then γi (x) → γ(x) uniformly for x
belonging to any compact subset of G.
Theorem 66. If y ∈ R, then γy (x) = eixy denes a character on R and every
character on R has this form. The map y 7→ γy is a homeomorphism and an
isomorphism of R onto R̂. If y ∈ R and f ∈ L1 (R), then

Z ∞
fˆ(γy ) = fˆ(y) = f (x)e−ixy , dx
−∞

the Fourier transform of f .

Theorem 67. If n ∈ Z, dene γn : T → T by γn (z) = z n . Then γn ∈ T̂ and

the map n 7→ γn is a homeomorphism and an isomorphism of Z onto T̂. If n ∈ Z
and f ∈ L1 (T), then

Z 2π
1
fˆ(γn ) = fˆ(n) ≡ f (eiθ )e−inθ dθ.
2π 0

Theorem 68. If G is a compact, Ĝ is discrete; if G is discrete, Ĝ is compact.

Theorem 69. If a ∈ T, dene γa : Z → T by γa (n) = an . Then γa ∈ Ẑ and

the map a 7→ γa is a homeomorphism and an isomorphism of T onto Ẑ. If a ∈ T
and f ∈ L1 (Z) = l1 (Z), then

∞
X
fˆ(γa ) = fˆ(a) = f (n)a−n .
n=−∞
 Bibliography
[1] John B. Conway - A course in Functional Analysis. Second Edition. Editorial:

Springer-Verlag 1985

[2] BERBERIAN, S.K. - Lectures in Functional Analysis and Operator Theory. Editorial:

Springer-Verlag, 1974

17