By
Dinesh J. Karia
Dedicated to:
All who critically see this Lecture Notes
Contents
Chapter 1. Introduction 1
1. A Brief History 2
2. Preliminaries 3
3. Examples 7
4. Group Algebras 14
5. New From Old 20
6. Homomorphisms 23
7. Adjoining Identity 24
8. Regular Elements 26
9. Topological Divisors of Zero 29
10. Spectrum 31
11. Gel’fand Mazur Theorems 33
12. Algebras Without Identity 34
Bibliography 36
iii
CHAPTER 1
Introduction
This course is designed in two folds. One is Banach Algebra and
the other is Operator Theory. The course will evolve along the following
guidelines listed in the syllabus of M-506.
Banach Algebras
Definition and examples of a Banach algebra – Adjoining identity
– Regular and singular elements – Topological divisors of zero – The
spectrum and spectral radius – Gel’fand-Mazur theorem – Complex
homomorphisms – Weak Topology.
Commutative Banach algebras – Maximal ideals – The radical –
Semisimplicity – Maximal ideal space– Maximal ideal space of C(X)
and A(D) (Functions continuous on the closed unit disc D and analytic
in the interior) – Wermer’s maximality theorem.
The Weierstrass approximation theorem – The Stone-Weierstrass
theorem – Involution in Banach algebra – The Gel’fand Naimark the-
orem for a commutative B ∗ -algebra.
Semisimple commutative Banach algebras – Homomorphisms of
commutative Banach algebra.
Closed ideals of C(X) – The Banach-Stone theorem – Stone-Čech
compactification.
Operator Theory
Normal, unitary, selfadjoint and positive operators – Orthogonal
projections – Invariant and reducing subspaces.
Spectra of operators – Properties of spectrum – Various parts of
spectrum – Spectral mapping theorem – Numerical range
Spectral analysis of selfadjoint and normal operators – Spectral
theorem of compact normal operators – Spectral theorem for selfad-
joint operators – Fuglede-Putnam theorem – Square root of a positive
operator – Polar decomposition of an operator
Matrix representation of an operator – Operator matrices – Direct
sum
1
2 1. INTRODUCTION
1. A Brief History
Functional Analysis has been extensively studied in the last century.
The theory aims at embarking the algebraic problems using the
Analysis. However, the work of Gel’fand in 1941 is considered
to be pioneering as far as the theory of Banach algebras is
concerned. Bonsall & Duncan [BD, p. 4] very much truly
state that “The name ‘Banach algebra’ is appropriate
only because, as a normed linear space, a Banach al-
gebra is a Banach space. However this name is too
firmly established in the literature to be changed
at this time. Given a free choice we should like
to call complete normed algebras ‘Gelfand alge-
bras’ in recognition of the distinguished pioneer-
ing work of I. M. Gel’fand in this field.” The His-
torical Notes in [Ru2, p. 373-374] makes it more
clear as to how and when the theory got the mo-
mentum. He points out that Banach’s book [Ban]
deals with the vector space structure only; he never
multiplied his objects. It was Wiener who, in 1932, dealt
with the multiplicative structure and the submultiplicativ-
ity of the norm on the Banach space of absolutely convergent
Fourier series. Stone used explicitly the ring structure of spaces
of continuous functions in his work in 1937 when he generalized the
Weierstrass approximation theorem, now known as Stone-Weierstrass
theorem. In fact, Nagumo (Japanese) first initiated the abstract study
of normed rings. But what really got this subject off the ground was
Gel’fand’s discovery (1941) of important role played the maximal ideals
of commutative algebra in his construction what is now known as the
Gel’fand transform.
The classic book [Ri] by Charles E. Rickart is like “Bible of Ba-
nach algebras.” The references [BD], [Br] [La], [Ru2] and [S] are also
worth going through. We shall mainly follow the books [S] by Simmons
and [La] by Larsen.
It was von Neumann who axiomatized Hilbert space in 1930 when
his paper laid the foundation stone for Operator Theory. Incidently,
he is the person who first thought and gave the idea of storing the
programs on computer while he was associated with the project of
2. PRELIMINARIES 3
2. Preliminaries
In this section we develop the basic definitions and results. We
also fix up notations and conventions here for the rest of our notes.
Notations: R and C will denote respectively the fields of real numbers
and complex numbers. In our discussion K will stand for either of them.
Let us first recall the algebraic structures you have studied so far.
2.1. Definition. Let G be a nonempty set with an associative binary
operation · with the following properties.
(1) There is e ∈ G such that xe = ex = x for all x ∈ G. (e turns out
to be unique and is called identity of G.)
(2) For each x ∈ G, there is y ∈ G such that xy = yx = e. (Given an
x such a y is unique and is called the inverse of x. It is customary
to denote inverse of x by x−1 .)
Then G is called a group. Further, if for all x, y ∈ G, xy = yx, then G
is said to be commutative or abelian.
If we add one more binary operation (with some properties) in a
group, then we get a ring.
2.2. Definition. Let (R, +) be an abelian group. Further, let · be
an associative binary operation on R such that x(y + z) = xy + xz
and (y + z)x = yx + zx for all x, y, z ∈ R. Then (R, +, ·) is called a
ring. These operations are called addition and multiplication. It is a
common practice to omit the dot for the ring-multiplication. That is,
xy stands for x · y.
(1) If, in a ring R, xy = yx for all x, y ∈ R, then R is called a
commutative ring.
(2) If there is an element e in a ring R satisfying xe = ex = x for
all x ∈ R, then such an e turns out to be unique and is called
the identity or unity of R. In such a case, R is called a ring
with identity or ring with unity or unital ring. Many a times the
identity of R is denoted by 1.
4 1. INTRODUCTION
3. Examples
First we note that (R, k · k) with the usual operations is a Banach
algebra over R and (C, k · k) with the usual operations is a Banach
algebra over C.
8 1. INTRODUCTION
Further let
`∞ = {x = (xi )∞
i=1 : kxk∞ = sup |xi | < ∞}. (3.2.2)
i=1,2,...
Proof. You already know that (`p , k · kp ) are Banach spaces for all
p ∈ [1, ∞]. So, we verify only the submultiplicativity of the norm
on A. Fix p ∈ [1, ∞). Let x = (xi ), y = (yi ) ∈ A. Assume first that
1 ≤ p < ∞. Clearly, the sequences are bounded as they form summable
series. Also,
p1 p1
∞ ∞
p p
|xi | ≤ Σ |xi | for all i ⇒ kxk∞ ≤ Σ |xi | = kxkp .
i=1 i=1
3. EXAMPLES 9
Hence,
p1
∞
p
kxykp = Σ |xi yi |
i=1
p1
∞
p
≤ Σ |kxk∞ yi |
i=1
p1
∞
p
= kxk∞ Σ |yi |
i=1
4. Group Algebras
Let G = {g1 , g2 , . . . , gn } be a finite group. Let A be the set of all
K-valued functions defined on G. An element x of A is denoted by
n
x = Σ x(gi )gi (4.0.1)
i=1
We also define
n
kxk1 = Σ |x(gi )| (4.0.6)
i=1
n n
(x ? y) ? z = ( Σ x(gi )gi ? Σ y(gj )gj ) ? z
i=1 j=1
n n
={Σ[ Σ x(gi )y(gj )]gk } ? Σ z(gm )gm
k=1 gi gj =gk m=1
n
= Σ{ Σ [ Σ x(gi )y(gj )]z(gm )}gp
p=1 gk gm =gp gi gj =gk
n
= Σ{ Σ x(gi )y(gj )z(gm )}gp . (4.1.1)
p=1 gi gj gm =gp
Also,
n n
x ? (y ? z) = x ? ( Σ y(gj )gj ? Σ z(gm )gm )
j=1 m=1
n n
= Σ x(gi )gi ? { Σ [ Σ y(gj )z(gm )]gk }
i=1 k=1 gj gm =gk
n
= Σ{ Σ x(gi )[ Σ y(gj )z(gm )]}gp
p=1 gi gk =gp gj gm =gk
n
= Σ{ Σ x(gi )y(gj )z(gm )}gp . (4.1.2)
p=1 gi gj gm =gp
= x ? y + x ? z. (4.1.3)
16 1. INTRODUCTION
Hence
n
(x ? y)(gk ) = Σ x(gi )y(gj ) = Σ x(gk gj−1 )y(gj ) (4.1.4)
gi gj =gk j=1
As a result,
n
kx ? yk1 = Σ |(x ? y)(gk )|
k=1
n n
= Σ | Σ x(gk gj−1 )y(gj )|
k=1 j=1
n n
≤ Σ { Σ |x(gk gj−1 )||y(gj )|}
k=1 j=1
n n
= Σ Σ {|x(gk gj−1 )||y(gj )|}
k=1 j=1
n n
= Σ Σ {|x(gk gj−1 )||y(gj )|}
j=1 k=1
n n
= Σ { Σ |x(gk gj−1 )||y(gj )|}
j=1 k=1
n n
= Σ {|y(gj )| Σ |x(gk gj−1 )|}
j=1 k=1
n
= Σ (|y(gj )|kxk1 )
j=1
n
= kxk1 Σ |y(gj )|
j=1
= kxk1 kyk1 .
∞ ∞
For x = Σ x(gi )gi , y = Σ x(gi )gi ∈ `1 (G), define
i=1 i=1
∞ ∞ ∞
x?y = Σ [ Σ x(gi )y(gj )]gk = Σ [ Σ x(gk gj−1 )y(gj )]gk . (4.2.2)
k=1 gi gj =gk k=1 j=1
∞
Σ |x(gk gj−1 )y(gj )| ≤ rkyk1 . (4.2.3)
j=1
∞
Thus, for each k, the series Σ x(gk gj−1 )y(gj ) converges in K. Putting
j=1
∞
gk gj−1 = gi , we have Σ x(gk gj−1 )y(gj ) = Σ x(gi )y(gj ). Thus the
j=1 gi gj =gk
18 1. INTRODUCTION
Hence,
∞ ∞ ∞
kx ? yk1 = Σ |(x ? y)(gk )| = Σ | Σ x(gk gj−1 )y(gj )|
k=1 k=1 j=1
∞ ∞ ∞ ∞
≤ Σ { Σ |x(gk gj−1 )||y(gj )|} = Σ Σ {|x(gk gj−1 )||y(gj )|}
k=1 j=1 k=1 j=1
∞ ∞ ∞ ∞
= Σ Σ {|x(gk gj−1 )||y(gj )|} = Σ { Σ |x(gk gj−1 )||y(gj )|}
j=1 k=1 j=1 k=1
∞ ∞ ∞
= Σ {|y(gj )| Σ |x(gk gj−1 )|} = Σ (|y(gj )|kxk1 )
j=1 k=1 j=1
∞
= kxk1 Σ |y(gj )| = kxk1 kyk1 .
j=1
For x = (xi )∞ ∞ 1
−∞ , y = (yi )−∞ ∈ ` (Z) we have
∞
x ? y(j) = Σ xi yj−i . (4.3.2)
i=−∞
Then with these operations and the norm à becomes a Banach algebra
and (A, k · k) becomes a subalgebra of Ã.
Proof. First we show that the sequences {}
5.4. Example. The following lists several examples of subalgebras and
ideals of different algebras. The proofs are straight verifications left to
the reader.
(1) Let A0 = (C[a, b], k · k∞ ) and An = (C n [a, b], k · k∞ ). Then each
An is a subalgebra of An−1 , n ∈ N. Also let x ∈ [a, b] be fixed.
5. NEW FROM OLD 21
Define
Inx = {f ∈ An : f (x) = 0}.
Then Inx is an ideal of An .
(2) Let A be an algebra and x ∈ A. Then
Ax = {ax : a ∈ A}
is a left ideal of A.
(3) Consider Kn with pointwise operations. For 1 ≤ k ≤ n, define
Ik = {x = (xi ) ∈ Kn : xk = 0}. Then Ik is an ideal of Kn .
(4) c00 is an ideal of `p , (1 ≤ p ≤ ∞).
(5) c00 is an ideal of c as well as of c0 .
(6) c0 is an ideal of c.
n
(7) I = {p = a0 + a1 x + a2 x2 + · · · + an xn ∈ P[0, 1] : Σ ai = 0} is
i=1
and ideal of P[0, 1].
n
(8) I = {p = a0 + a1 x + a2 x2 + · · · + an xn ∈ P[0, 1] : Σ (−1)i ai = 0}
i=1
is and ideal of P[0, 1].
5.5. Theorem. Let A be an algebra with identity. Let x ∈ A. Then
the following are equivalent.
(1) x has a left inverse.
(2) x does not belong to any left ideal of A.
Proof. Assume (1). Let y be a left inverse of x. Suppose I is a proper
left ideal of A such that x ∈ I. Then 1 = yx ∈ I. Hence for any z ∈ A,
we have z = z1 = (zy)x ∈ I, giving A ⊂ I, a contradiction. Thus (2)
follows.
Assume (2). Clearly, x = 1x ∈ Ax and Ax is a left ideal of A. So, Ax
cannot be a proper ideal. That is, A = Ax. So, 1 ∈ Ax. As a result,
1 = yx for some y ∈ A. Thus x is left regular.
The following theorem can be proved similarly.
5.6. Theorem. Let A be an algebra with identity. Let x ∈ A. Then
the following are equivalent.
(1) x has a right inverse.
(2) x does not belong to any right ideal of A.
5.7. Corollary. A proper left or right ideal I of an algebra A cannot
contain any invertible element of A. In particular, it cannot contain 1.
22 1. INTRODUCTION
5.8. Example. Let A = K×K with pointwise addition and scalar mul-
tiplication. For (x1 , y1 ), (x2 , y2 ) ∈ K2 define (x1 , y1 )(x2 , y2 ) = (x1 x2 +
x1 y2 + x2 y1 , y1 y2 ) and k(x1 , y1 )k1 = |x1 | + |y1 |. Then (A, k · k1 ) is a
Banach algebra. Now let B = {(x, 0) : x ∈ K}. Then the following are
easily seen.
(1) A has unit element (0, 1).
(2) B has unit element (1, 0).
Thus A and B are both unital algebras and B is a subalgebra of A.
However, since unit of A is not in B, B is not a unital subalgebra of A.
Also note here that B is an ideal of A and B has identity. But since it
is proper, it cannot contain the identity of A.
5.9. Example. Let A = K×K with pointwise addition and scalar mul-
tiplication. For (x1 , y1 ), (x2 , y2 ) ∈ K2 define (x1 , y1 )(x2 , y2 ) = (x1 x2 , 0)
and k(x1 , y1 )k1 = |x1 | + |y1 |. Then (A, k · k1 ) is a Banach algebra. Now
let B = {(x, 0) : x ∈ K}. Then the following are easily seen.
(1) A does not have a unit element.
(2) B has unit element (1, 0).
Thus A is not a unital algebra but B is a unital algebra. Also, B is a
subalgebra of A. However, B is not a unital subalgebra of A. Why?
5.10. Example. Let M2 denote the algebra of all 2 × 2 matrices over
a a
K. Define B = : a ∈ K . Then B is a subalgebra of M2 .
a a
Here M2 and B both are unital but B is not a unital subalgebra of
M2 .
ka ma
5.11. Exercise. For fixed k, m ∈ K, let C = :a∈K .
ka ma
Show that C is a subalgebra of the algebra M2 . Obtain conditions on
k, m so that the algebra C becomes a unital algebra. In any case, C is
not a unital subalgebra of M2 .
6. Homomorphisms
6.1. Definitions. Let A, B be two algebras over the same field K. A
linear mapping ϕ : A → B is said to be an algebra homomorphism from
A to B if ϕ(xy) = ϕ(x)ϕ(y), for all x, y ∈ A. The set ker(ϕ) = {x ∈
A : ϕ(x) = 0} is called the kernel of ϕ. Further, if ϕ is one-one, then
it is called an algebra isomorphism. If ϕ is one-one and onto, then it
is called an onto algebra isomorphism and in this case, we say that A
is algebraically isomorphic to B. Further, suppose that A and B are
normed algebras. If the algebra isomorphism ϕ is a homeomorphism
between, A and ϕ(A), then we say that ϕ is a topological isomorphism
and A and ϕ(A) are topologically isomorphic. Further if kϕ(x)k = kxk
for all x ∈ A, then ϕ is called an isometric isomorphism and A and
ϕ(A) are called isometric isomorphic.
6.2. Remark. While dealing with more than one algebras, we shall
assume that they are over a same field K, unless otherwise stated.
7. Adjoining Identity
In general, an algebra need not have identity. In that case, we can
obtain a bigger algebra with identity containing the given algebra as a
subalgebra.
7.1. Theorem. [La, p.10] Let A be a normed algebra without identity.
Define Ae = A × K. For (x, λ), (y, µ) ∈ Ae and k ∈ K, define
(x, λ) + (y, µ) = (x + y, λ + µ).
7. ADJOINING IDENTITY 25
8. Regular Elements
The set of all regular elements play an important role in the theory
of Banach algebras.
8.1. Theorem. Let A be a Banach algebra with identity. Then the
following hold
8. REGULAR ELEMENTS 27
(1) G is a group.
∞
(2) If x ∈ A and if k1 − xk < 1, then x ∈ G and x−1 = 1+ Σ (1−x)n .
n=1
(3) G is open in A.
(4) The mapping x ∈ G 7→ x−1 ∈ G is a homeomorphism.
y − xy = (1 − x)y
∞
= (1 − x) Σ (1 − x)n
n=0
∞
= Σ (1 − x)n
n=1
∞
= 1 + Σ (1 − x)n − 1
n=1
∞
= Σ (1 − x)n − 1
n=0
=y−1
Thus xy = 1. Also,
y − yx = y(1 − x)
∞
= Σ (1 − x)n (1 − x)
n=0
∞
= Σ (1 − x)n
n=1
∞
= 1 + Σ (1 − x)n − 1
n=1
∞
= Σ (1 − x)n − 1
n=0
=y−1
Thus yx = 1.
28 1. INTRODUCTION
−1
(3) Let x0 ∈ G and kx − x0 k < kx−1
0 k . Then
−1
k1 − x−1 −1 −1 −1 −1
0 xk = kx0 (x0 − x)k ≤ kx0 kk(x0 − x)k < kx0 kkx0 k = 1.
Hence x−1 −1 −1
0 x ∈ G. Thus x0 , x0 x ∈ G. So, x = x0 (x0 x) ∈ G.
(4) Let x0 ∈ G, we show that(the function x 7→)x−1 is continuous at
ε 1
x0 . For an ε > 0, let δ = min ,
−1 2 2kx−1 k
. Let kx − x0 k < δ.
2kx0 k 0
Then
1 1 1
kx−1 −1 −1 −1
0 x − 1k = kx0 (x − x0 )k ≤ kx0 kkx − x0 k < kx0 k −1 = = 1− .
2kx0 k 2 2
Hence,
1 − kx−1
0 x − 1k > 1/2.
So,
1
< 2. (8.1.1)
1− kx−1
0 x − 1k
∞
Consequently, x−1 x0 = (x−1
0 x)
−1
= Σ (1 − x−1 n
0 x) . Thus
n=0
kx −1
− x−1
0 k = k(x−1
0 x − 1)x−1
0 k
≤ kx−1 −1
0 x − 1kkx0 k
∞
= kx−1 −1 n
0 kk Σ (1 − x0 x) − 1k
n=0
∞
≤ kx−1 −1 n
0 kk Σ (1 − x0 x) k
n=1
∞ n
≤ kx−1 −1
0 k Σ k1 − x0 xk
n=1
−1
kx0 kk1 −x−1
0 xk
= −1
1 − k(1 − x0 x)k
= 2kx−1 −1
0 kkx0 (x0 − x)k
2
≤ 2kx−1
0 k kx0 − xk
2 2 ε
< 2kx−1 −1
0 k δ ≤ 2kx0 k 2 < ε.
2kx−1
0 k
1 xx−1 −1
nk − xnk xnk 1 (x − xnk )x−1
nk
= + = +
kx−1
nk k kx −1
nk k kx −1
nk k kx −1
nk k
1
= −1 + (x − xnk )zk → 0
kxnk k
Thus x is a topological divisor of zero.
9.3. Corollary. Let x be a singular element in a Banach algebra A
and {xn } be a sequence of invertible elements such that xn → x. Then
{x−1
n } is unbounded.
10. Spectrum
As we know the eigenvalues of a square matrix plays important
role in investigating the matrix. Let x be an n × n matrix and e denote
the n×n matrix. Then λ is an eigenvalue of x iff x−λe is not invertible.
We generalize this concept to the setup of an algebra.
10.1. Definition. Let A be an algebra with identity 1. For x ∈ A, we
define the spectrum of x to be the set
spA (x) = {λ ∈ K : x − λ1 is not invertible in A}.
It is also denoted by σA (x). The complement of the spectrum of x is
called the resolvent set of x and is denoted by ρA (x).
Note here that the computation of the spectrum of an element
depends on the algebra. As earlier, if we consider the function f (z) = z,
then f is not invertible in A(D), but is invertible in C(Γ). Thus 0
belongs to the spectrum of f calculated with respect to A(D) but 0
does not belong to the spectrum of f calculated with respect to C(Γ).
However, we will drop A from the notation spA (x) and ρA (x), whenever
there is no room for confusion.
10.2. Lemma. Let A be an algebra with identity 1 and x ∈ A. Then
sp(x) is closed and bounded subset of K. That is, sp(x) is a compact
subset of x.
Proof. Let λ ∈ K such that |λ| > kxk. Then k λx k < 1. Hence, 1 − λx
is invertible. Consequently, (x − λ1) = (−gl)(1 − λx ) is invertible. That
is λ ∈ sp(x) ⇒ λ ≤ kxk. Also, consider the continuous map defined as
λ ∈ K 7→ x − λ1 ∈ A. Then sp(x) is the inverse image of the closed
set A \ G. Thus sp(x) must be closed.
10.3. Remark.
(1) The completeness of A cannot be dropped from the above Lemma.
Indeed, for any nonconstant polynomial p ∈ P[0, 1], sp(x) = K.
(2) In general, for a real Banach algebra A, spectrums of some of its
elements may be empty. Indeed, considering C as a real algebra,
we see that the invertible elements are precisely, the non-zero
elements. But i − λ 6= 0 for any λ ∈ R. Hence the spectrum of
i is empty. However, the complex algebras are very nice! They
assure that the spectrums of their elements are nonempty.
32 1. INTRODUCTION
36