442C Banach algebras 200910

1 Introduction to Banach algebras

1.1 Denitions and examples
Let us adopt the convention that all vector spaces and Banach spaces are
over the eld of complex numbers.
1.1.1 Denition. A Banach algebra is a vector space A such that
(i). A is an algebra: it is equipped with an associative product (a, b) ab
which is linear in each variable,
(ii). A is a Banach space: it has a norm | | with respect to which it is
complete, and
(iii). A is a normed algebra: we have |ab| |a| |b| for a, b A.
If the product is commutative, so that ab = ba for all a, b A, then we say
that A is an abelian Banach algebra.
1.1.2 Remark. The inequality |ab| |a| |b| ensures that the product is
continuous as a map A A A.
1.1.3 Examples. (i). Let X be a topological space. We write BC(X) for
the set of bounded continuous functions X C. Recall from [FA1.7.2]
that BC(X) is a Banach space under the pointwise vector space oper-
ations and the uniform norm, which is given by
|f| = sup
xX
[f(x)[, f BC(X).
The pointwise product
(fg)(x) = f(x)g(x), f, g BC(X), x X
turns BC(X) into an abelian Banach algebra. Indeed, the product is
clearly commutative and associative, it is linear in f and g, and
|fg| = sup
xX
[f(x)[ [g(x)[ sup
x
1
X
[f(x
1
)[ sup
x
2
X
[g(x
2
)[ = |f| |g|.
If X is a compact space, then every continuous function X C is
bounded. For this reason, we will write C(X) instead of BC(X) if X
is compact.
1
(ii). If A is a Banach algebra, a subalgebra of A is a linear subspace B A
such that a, b B = ab B. If B is a closed subalgebra of a
Banach algebra A, then B is complete, so it is a Banach algebra (under
the same operations and norm as A). We then say that B is a Banach
subalgebra of A.
(iii). Let D = z C: [z[ < 1 and D = z C: [z[ 1. The disc algebra
is the following Banach subalgebra of C(D):
A(D) = f C(D): f is analytic on D.
(iv). The set C
0
(R) of continuous functions f : R C such that
lim
x
f(x) = lim
x
f(x) = 0.
This is a Banach subalgebra of BC(R).
(v). Let
1
(Z) denote the vector space of complex sequences (a
n
)
nZ
in-
dexed by Z such that |a| =

nZ
[a
n
[ < . This is a Banach space
by [FA1.7.10]. We dene a product such that if a = (a
n
)
nZ
and
b = (b
n
)
nZ
are in
1
(Z) then the nth entry of a b is
(a b)
n
=

mZ
a
m
b
nm
.
This series is absolutely convergent, and a b
1
(Z), since
|a b| =

n
[(a b)
n
[ =

m
a
m
b
nm

m,n
[a
m
[ [b
nm
[ =

m
[a
m
[

n
[b
mn
[ = |a| |b| < .
It is an exercise to show that is commutative, associative and linear
in each variable, so it turns
1
(Z) into an abelian Banach algebra.
(vi). If X is a Banach space, let B(X) denote the set of all bounded linear
operators T : X X with the operator norm
|T| = sup
xX, x1
|Tx|.
By [FA3.3], B(X) is a Banach space. Dene a product on B(X) by
ST = S T. This is clearly associative and bilinear, and if x X with
|x| 1 then
|(ST)x| = |S(Tx)| |S| |Tx| |S| |T|
so |ST| |S| |T|. Hence B(X) is a Banach algebra. If dimX > 1
then B(X) is not abelian.
2
1.2 Invertibility
1.2.1 Denition. A Banach algebra A is unital if A contains an identity
element of norm 1; that is, an element 1 A such that 1a = a1 = a for all
a A, and |1| = 1. We call 1 the unit of A. We sometimes write 1 = 1
A
to
make it clear that 1 is the unit of A.
If A is a unital Banach algebra and B A, we say that B is a unital
subalgebra of A if B is a subalgebra of A which contains the unit of A.
1.2.2 Examples. Of the Banach algebras in Example 1.1.3, only C
0
(R) is
non-unital. Indeed, it is easy to see that no f C
0
(R) is an identity element
for C
0
(R). On the other hand the constant function taking the value 1 is the
unit for BC(X), and A(D) is a unital subalgebra of C(D). Also, the identity
operator I : X X, x x is the unit for B(X), and is it not hard to check
that the sequence (
n,0
)
nZ
is the unit for
1
(Z).
1.2.3 Remarks. (i). An algebra can have at most one identity element.
(ii). If (A, | |) is a non-zero Banach algebra with an identity element then
we can dene an norm [ [ on A under which it is a unital Banach
algebra such that [ [ is equivalent to | |, meaning that there are
constants m, M 0 such that
m[a[ |a| M[a[ for all a A.
For example, we could take [a[ = |L
a
: A A| where L
a
is the linear
operator L
a
(b) = ab for a, b A.
1.2.4 Denition. Let A be a Banach algebra with unit 1. An element a A
is invertible if ab = 1 = ba for some b A. Its easy to see that b is then
unique; we call b the inverse of a and write b = a
1
.
We write Inv A for the set of invertible elements of A.
1.2.5 Remarks. (i). Inv A forms a group under multiplication.
(ii). If a A is left invertible and right invertible so that ba = 1 and ac = 1
for some b, c A, then a is invertible.
(iii). If a = bc = cb then a is invertible if and only if b and c are invertible.
It follows by induction that if b
1
, . . . , b
n
are commuting elements of A
(meaning that b
i
b
j
= b
j
b
i
for 1 i, j n) then b
1
b
2
. . . b
n
is invertible
if and only if b
1
, . . . , b
n
are all invertible.
(iv). The commutativity hypothesis is essential in (iii). For example, if
(e
n
)
n1
is an orthonormal basis of a Hilbert space H and S B(H) is
dened by Se
n
= e
n+1
, n 1 and S

is the adjoint of S (see [FA4.18])

then S

S is invertible although S

and S are not.

3
1.2.6 Examples. (i). If X is a compact topological space then
Inv C(X) = f C(X): f(x) ,= 0 for all x X.
Indeed, if f(x) ,= 0 for all x X then we can dene g : X C,
x f(x)
1
. The function g is then continuous [why?] with fg = 1.
Conversely, if x X and f(x) = 0 then fg(x) = f(x)g(x) = 0 so
fg ,= 1 for all g C(X), so f is not invertible.
(ii). If X is a Banach space then
Inv B(X)
_
T B(X): ker T = 0
_
.
Indeed, if ker T ,= 0 then T is not injective, so cannot be invertible.
If X is nite-dimensional and ker T = 0 then by linear algebra,
T is surjective, so T is an invertible linear map. Since X is nite-
dimensional, the linear map T
1
is bounded, so T is invertible in B(X).
Hence
Inv B(X) =
_
T B(X): ker T = 0
_
if dimX < .
On the other hand, if X is innite-dimensional then we generally have
Inv B(X)
_
T B(X): ker T = 0
_
. For example, let X = H be
an innite-dimensional Hilbert space with orthonormal basis (e
n
)
n1
.
Consider the operator T B(H) dened by
Te
n
=
1
n
e
n
, n 1.
It is easy to see that ker T = 0. However, T is not invertible. Indeed,
if S B(H) with ST = I then Se
n
= S(nTe
n
) = nSTe
n
= ne
n
, so
|Se
n
| = n as n
and so S is not bounded, which is a contradiction.
1.2.7 Theorem. Let A be a Banach algebra with unit 1. If a A with
|a| < 1 then 1 a Inv A and
(1 a)
1
=

n=0
a
n
.
Proof. Since |a
n
| |a|
n
and |a| < 1, the series

n=0
a
n
is absolutely
convergent and so convergent by [FA1.7.8], say to b A. Let b
n
be the nth
partial sum of this series and note that
b
n
(1 a) = (1 a)b
n
= (1 a)(1 + a + a
2
+ + a
n
) = 1 a
n+1
1
as n . So b(1 a) = (1 a)b = 1 and so b = (1 a)
1
.
4
1.2.8 Corollary. Inv A is an open subset of A.
Proof. Let a Inv A and let r
a
= |a
1
|
1
. We claim that the open ball
B(a, r
a
) = b A: |a b| < r
a
is contained in Inv A; for if b B(a, r
a
)
then |a b| < r
a
and
b = (a (a b))a
1
a = (1 (a b)a
1
)a.
Since |(a b)a
1
| < r
a
|a
1
| < 1, the element 1 (a b)a
1
is invertible
by Theorem 1.2.7. Hence b is the product of two invertible elements, so is
invertible. This shows that every element of Inv A may be surrounded by an
open ball which is contained in Inv A, hence Inv A is open.
1.2.9 Corollary. The map : Inv A Inv A, a a
1
is a homeomor-
phism.
Proof. Since (a
1
)
1
= a, the map is a bijection with =
1
. So we only
need to show that is continuous.
If a Inv A and b Inv A with |a b| <
1
2
|a
1
|
1
then using the
triangle inequality and the identity
a
1
b
1
= a
1
(b a)b
1
()
we have
|b
1
| |a
1
b
1
|+|a
1
| |a
1
| |ab| |b
1
|+|a
1
|
1
2
|b
1
|+|a
1
|,
so |b
1
| 2|a
1
|. Using () again, we have
|(a) (b)| = |a
1
b
1
| |a
1
| |a b| |b
1
| < 2|a
1
|
2
|a b|,
which shows that is continuous at a.
1.3 The spectrum
1.3.1 Denition. Let A be a unital Banach algebra and let a A. The
spectrum of a in A is
(a) =
A
(a) = C: 1 a , Inv A.
We will often write instead of 1 for C.
5
1.3.2 Examples. (i). We have (1) = for any C.
(ii). Let X be a compact topological space. If f C(X) then
(f) = f(X) = f(x): x X.
Indeed,
(f) 1 f , Inv C(X)
(1 f)(x) = 0 for some x X, by Example 1.2.6(i)
= f(x) for some x X
f(X).
(iii). If X is a nite-dimensional Banach space and T B(X) then
(T) = C: is an eigenvalue of T.
Indeed,
(T) T , Inv B(X)
ker(I T) ,= 0 by Example 1.2.6(ii)
(I T)(x) = 0 for some nonzero x X
Tx = x for some nonzero x X
is an eigenvalue of T.
If X is an innite-dimensional Banach space, then the same argument
shows that (T) contains the eigenvalues of T, but generally this in-
clusion is strict.
We will need the following algebraic fact later on.
1.3.3 Proposition. Let A be a unital Banach algebra and let a, b A.
(i). If 1 ab Inv A then 1 ba Inv A, and
(1 ba)
1
= 1 + b(1 ab)
1
a.
(ii). (ab) 0 = (ba) 0.
Proof. Exercise.
6
To show that the spectrum is always non-empty, we will use a vector-
valued version of Liouvilles theorem:
1.3.4 Lemma. Let X be a Banach space and suppose that f : C X is
an entire function in the sense that
f()f()

converges in X as , for
every C. If f is bounded then f is constant.
Proof. Given a continuous linear functional X

, let g = f : C C.
Since
g()g()

= (
f()f()

) and [g()[ |g| |f()|, the function g is entire

and bounded. By Liouvilles theorem it is constant, so (f()) = (f())
for all X

and , C. By the Hahn-Banach theorem (see [FA3.8]),

f() = f() for all , C. So f is constant.
1.3.5 Theorem. Let A be a unital Banach algebra. If a A then (a) is a
non-empty compact subset of C with (a) C: [[ |a|.
Proof. The map i : C A, a is continuous and
(a) = C: i() , Inv A = C i
1
(Inv A).
Since Inv A is open by Corollary 1.2.8 and i is continuous, i
1
(Inv A) is open
and so its complement (a) is closed.
If [[ > |a| then a = (1
1
a) and |
1
a| = [[
1
|a| < 1 so a is
invertible by Theorem 1.2.7, so , (a). Hence (a) C: [[ |a|.
In particular, (a) is bounded as well as closed, so (a) is a compact subset
of C.
Finally, we must show that (a) ,= . If (a) = then the map
R: C A, ( a)
1
is well-dened. It not hard to show using () that
R() R()

= R()R() for , C with ,= .
Corollary 1.2.9 shows that R is continuous, so we conclude that R is an entire
function (with derivative R

() = R()
2
).
Now |R()| = |( a)
1
| = [[
1
|(1
1
a)
1
| and 1
1
a 1 as
[[ so, by Corollary 1.2.9, (1
1
a)
1
1. Hence |R()| 0 as
[[ .
Hence R is a bounded entire function, so it is constant by Lemma 1.3.4;
since R() 0 as [[ we have R() = 0 for all C. This is a
contradiction since R() is invertible for any C.
7
The next result says that C is essentially the only unital Banach algebra
which is also a eld.
1.3.6 Corollary (The Gelfand-Mazur theorem). If A is a unital Banach
algebra in which every non-zero element is invertible then A = C1
A
.
Proof. Let a A. Since (a) ,= there is some (a). Now 1a , Inv A,
so 1 a = 0 and a = 1 C1.
1.3.7 Denition. If a is an element of a unital Banach algebra and p C[z]
is a complex polynomial, say p =
0
+
1
z + +
n
z
n
where
0
,
1
, . . . ,
n
are complex numbers, then we write
p(a) =
0
1 +
1
a + +
n
a
n
.
1.3.8 Theorem (The spectral mapping theorem for polynomials). If p is a
complex polynomial and a is an element of a unital Banach algebra then
(p(a)) = p((a)) = p(): (a).
Proof. If p is a constant then this immediate since (1) = . Suppose
that n = deg p 1 and let C. Since C is algebraically closed, we can
write
p = C(
1
z) . . . (
n
z)
for some C,
1
, . . . ,
n
C. Then
p(a) = C(
1
a) . . . (
n
a)
and the factors
i
a all commute. So
(p(a)) p(a) is not invertible
some
i
a is not invertible (by Remark 1.2.5(iii))
some
i
is in (a)
(a) contains a root of p
= p() for some (a).
1.3.9 Denition. Let A be a unital Banach algebra. The spectral radius of
an element a A is
r(a) = r
A
(x) = sup

A
(a)
[[.
1.3.10 Remark. We have r(a) |a| by Theorem 1.3.5.
8
1.3.11 Examples. (i). If X is a compact topological space and f C(X),
then
r(f) = sup

A
(f)
[[ = sup
f(X)
[[ = |f|.
(ii). To see that strict inequality is possible, take X = C
2
with the usual
Hilbert space norm and let T B(X) be the operator with matrix
_
0 1
0 0
_
. Since det(T I) =
2
, the only eigenvalue of T is 0 and so
(T) = 0 by Example 1.3.2(iii). Hence r(T) = 0 < 1 = |T|.
1.3.12 Theorem (The spectral radius formula). The spectral radius of an
element of a unital Banach algebra is given by
r(a) = lim
n
|a
n
|
1/n
= inf
n1
|a
n
|
1/n
.
Proof. If (a) and n 1 then
n
(a
n
) by Theorem 1.3.8. So [[
n

|a
n
| by Theorem 1.3.5, hence [[ |a
n
|
1/n
and so r(a) inf
n1
|a
n
|
1/n
.
Consider the function
S: C: [[ < 1/r(a) A, (1 a)
1
.
Observe that for [[ < 1/r(a) we have r(a) = [[r(a) < 1 by Theorem 1.3.8,
so 1 a is invertible and S() is well-dened. We can argue as in the
proof of Theorem 1.3.5 to see that S is holomorphic. By Theorem 1.2.7
we have S() =

n=0

n
a
n
for [[ < 1/|a|. If A

with || = 1
then the complex-valued function f = S is given by the power series
f() =

n=0
(a
n
)
n
for [[ < 1/|a|. Moreover, f is holomorphic for
[[ < 1/r(a), so this power series converges to f() for [[ < 1/r(a). Hence
for R > r(a) we have
(a
n
) =
1
2i
_
||=1/R
f()

n+1
d
and we obtain the estimate
[(a
n
)[
1
2

2
R
R
n+1
sup
||=1/R
[(S())[ R
n
M(R)
where M(R) = sup
||=1/R
|S()|, which is nite by the continuity of S on
the compact set C: [[ = 1/R. Since S() ,= 0 for any in the domain
of S, we have M(R) > 0. Hence
limsup
n1
|a
n
|
1/n
limsup
n1
RM(R)
1/n
= R
9
whenever R > r(a). We conclude that
r(a) inf
n1
|a
n
|
1/n
liminf
n1
|a
n
|
1/n
limsup
n1
|a
n
|
1/n
r(a)
and the result follows.
1.3.13 Corollary. If A is a unital Banach algebra and B is a closed unital
subalgebra of A then r
A
(b) = r
B
(b) for all b B.
Proof. The norm of an element of B is the same whether we measure it in B
or in A. By the spectral radius formula, r
A
(b) = lim
n1
|b
n
|
1/n
= r
B
(b).
While the spectral radius of an element of a Banach algebra does not
depend if we compute it in a subalgebra, the spectrum itself can change. We
explore this in the next few results.
Suppose that A is a unital Banach algebra and B is a unital Banach
subalgebra of A. If an element of b is invertible in B, then it is invertible
in A; so Inv B B Inv A. However, this inclusion may be strict, as the
following example shows.
1.3.14 Example. Recall that A(D) is the disc algebra of continuous func-
tions D C which are holomorphic on D. Note that, by the maximum
modulus principle, sup
zD
[f(z)[ = sup
[0,2)
[f(e
i
)[. Hence |f| = |f[
T
|,
and the map A(D) C(T), f f[
T
is a unital isometric isomorphism. So
we may identify A(D) with A(T) = f[
T
: f A(D), which is a closed unital
subalgebra of C(T).
Consider the function f(z) = z for z D. This is not invertible in A(D)
since f(0) = 0. Hence f[
T
is not invertible in A(T). However, f is invertible
in C(T) with inverse g : e
i
e
i
. So Inv A(T) A(T) Inv C(T).
1.3.15 Denition. If A is a unital Banach algebra then a subalgebra B A
with 1 B is said to be inverse-closed if Inv B = B Inv A; that is, if every
b B which is invertible in A also has b
1
B.
Clearly, if B is an inverse-closed unital subalgebra of Athen
B
(b) =
A
(b)
for all b B.
If K is a non-empty compact subset of C then exactly one of the connected
components of C K is unbounded. The bounded components of C K are
called the holes of K. If A is a unital Banach algebra and a A, let us write
R
A
(a) = C
A
(a) = C: a Inv A.
This is sometimes called the resolvent set of a. Note that the bounded
connected components of R
A
(a) are precisely the holes of
A
(a).
10
1.3.16 Theorem. Let B be a closed subalgebra of a unital Banach algebra A
with 1 B. If b B then
B
(b) is the union of
A
(b) with zero or more of
the holes of
A
(b). In particular, if
A
(b) has no holes then
B
(b) =
A
(b).
Proof. Since Inv B Inv A we have R
B
(b) R
A
(b), and so
A
(b)
B
(b),
for each b B. We claim that R
B
(b) is a relatively clopen subset of R
A
(b).
Since
B
(b) is closed by Theorem 1.3.5, R
B
(b) is open. The map
i : R
A
(b) A, ( b)
1
is continuous by Corollary 1.2.9, and
R
B
(b) = R
A
(b): i() = ( b)
1
B = i
1
(B).
Since B is closed, R
B
(b) is closed.
If G is a connected component of R
A
(b) then G R
B
(b) is either or
G. For otherwise, since R
B
(b) is clopen, G R
B
(b) and G R
B
(b) would be
proper clopen subsets of the connected set G, which is impossible. If G is
the unbounded component of R
A
(b) then, since
B
(b) is bounded, we must
have G
B
(b) = . The bounded components of R
A
(b) are precisely holes
of
A
(b). Hence

B
(b) =
A
(b)
_
G a hole of
A
(b): G
B
(b) ,= .
If
A
(b) has no holes then this reduces to
B
(b) =
A
(b).
1.3.17 Denition. Let A be a Banach algebra. If S A then the commu-
tant of S in A is
S

= a A: ab = ba for all b S.
The bicommutant of S in A is S

= (S

.
A set S A is commutative if ab = ba for all a, b S. Hence S is
commutative if and only if S S

.
1.3.18 Lemma. Let A be a Banach algebra. If T S A, then T

.
Moreover, S S

and S

= S

.
Proof. Exercise.
1.3.19 Proposition. Let A be a unital Banach algebra and let S A.
(i). S

is a closed, inverse-closed unital subalgebra of A.

(ii). If S is commutative then so is B = S

, and
B
(b) =
A
(b) for all
b B.
11
Proof. (i) Since multiplication is continuous on A, it is easy to see that the
commutant S

is closed. Clearly 1 S

, and using the linearity of multi-

plication shows that S

is a vector subspace of A, and it is a subalgebra by

associativity. If b S

Inv A then bc = cb for all c S, so cb

1
= b
1
c for
all c S, so b
1
S

and S

is inverse-closed.
(ii) We have S S

, so S

and S

by Lemma 1.3.18. Hence

B = S

is commutative. Moreover, B is an inverse-closed subalgebra of A

by (i), so
B
(b) =
A
(b) for all b B.
1.3.20 Denition. Let A be a Banach algebra without an identity element.
The unitisation of A is the Banach algebra

A whose underlying vector space
is AC with the product (a, )(b, ) = (ab+b+a, ) and norm|(a, )| =
|a| + [[ for a, b A and , C. Note that

A is then a unital Banach
algebra containing A (or, more precisely, A 0).
1.3.21 Denition. If A has no identity element and a A, then we dene

A
(a) =
e
A
(a). In this case we have 0 (A) for all a A.
1.3.22 Remark. With this denition, many of the important theorems
above apply to non-unital Banach algebras, simply by considering

A instead
of A. In particular, it is easy to check that non-unital versions of Theo-
rems 1.3.5, 1.3.8 and 1.3.12 hold.
1.4 Quotients of Banach spaces
Recall that if K is a subspace of a complex vector space X, then the quotient
vector space X/K is given by
X/K = x + K: x X
with scalar multiplication (x + K) = x + K, C, x X
and vector addition (x + K) + (y + K) = (x + y) + K, x, y X, C.
The zero vector in X/K is 0 + K = K.
1.4.1 Denition. If K is a closed subspace of a Banach space X then the
quotient Banach space X/K is the vector space X/K equipped with the
quotient norm, dened by
|x + K| = inf
kK
|x + k|.
1.4.2 Proposition. Let X be a Banach space and let K be a closed vector
subspace of X. The quotient norm is a norm on the vector space X/K, with
respect to which X/K is complete. Hence the quotient Banach space X/K is
a Banach space.
12
Proof. To see that the quotient norm is a norm, observe that:
|x + K| 0 with equality if and only if inf
kK
|x + k| = 0, which is
equivalent to x being in the closure of K; since K is closed, this means
that x K so x + K = K, the zero vector of X/K.
If C with ,= 0 then
|(x + K)| = inf
kK
|x + k| = [[ inf
kK
|x +
1
k|
= [[ inf
k

K
|x + k| = [[ |x + K|.
The triangle inequality holds since K = s + t : s, t K and so
|(x + K) + (y + K)| = |x + y + K| = inf
kK
|x + y + k|
= inf
s,tK
|x + y + s + t|
inf
sK
|x + s| + inf
tK
|y + t|
= |x + K| +|y + K|.
It remains to show that X/K is complete in the quotient norm. For any
x X, it is not hard to see that:
(i) if > 0 then there exists k K such that |x+k| < |x+K| +; and
(ii) |x + K| |x| (since 0 K).
Let x
j
X with

j=1
|x
j
+ K| < . From observation (i), it follows
that there exist k
j
K with

j=1
|x
j
+k
j
| < , so by [FA1.7.8] the series

j=1
x
j
+ k
j
converges in X, say to s X. By observation (ii),
_
_
_s +K
_
n

j=1
x
j
+K
__
_
_ =
_
_
_
_
s
n

j=1
x
j
+k
j
_
+K
_
_
_
_
_
_s
n

j=1
x
j
+k
j
_
_
_ 0
as n , so

j=1
x
j
+ k
j
converges to s + K. This shows that every ab-
solutely convergent series in X/K is convergent with respect to the quotient
norm, so X/K is complete by [FA1.7.8].
13
1.5 Ideals, quotients and homomorphisms of Banach
algebras
1.5.1 Denition. An ideal of a Banach algebra A is a vector subspace I
of A such that for all x I and a A we have ax I and xa I.
1.5.2 Denition. Let I be a closed ideal of a Banach algebra A. The
quotient Banach algebra A/I is the quotient Banach space A/I equipped
with the product (a + I)(b + I) = ab + I for a, b I.
1.5.3 Theorem. If I is a closed ideal of a Banach algebra A then A/I is a
Banach algebra. If A is abelian then so is A/I. If A is unital then so is A/I,
and 1
A/I
= 1
A
+ I.
Proof. We saw in Proposition 1.4.2 that A/I is a Banach space. Just as
for quotient rings, the product is well-dened, since if a
1
+ I = a
2
+ I and
b
1
+I = b
2
+I then a
1
a
2
I and b
1
b
2
I, so a
1
(b
1
b
2
)+(a
1
a
2
)b
2
I
and so
(a
1
b
1
+I) (a
2
b
2
+I) = a
1
b
1
a
2
b
2
+I = a
1
(b
1
b
2
) + (a
1
a
2
)b
2
+I = I,
hence a
1
b
1
+I = a
2
b
2
+I. It is easy to see that this product is linear in each
variable.
Let a, b A. We have
|a + I| |b + I| = inf
y,zI
|a + y| |b + z|
inf
y,zI
|(a + y)(b + z)| (by 1.1.1(iii) in A)
= inf
y,zI
|ab + (az + yb + yz)|
inf
xI
|ab + x| (since az + by + yz I for all y, z I)
= |ab + I| = |(a + I)(b + I)|.
Hence the inequality 1.1.1(iii) holds in A/I, and we have shown that A/I is
a Banach algebra.
If A is abelian then (a +I)(b +I) = ab +I = ba +I = (b +I)(a +I) for
all a, b A, so A/I is abelian. The proof of the nal statement about units
is left as an exercise.
1.5.4 Denition. A proper ideal of a Banach algebra A an ideal of A which
is not equal to A. A maximal ideal of A is a proper ideal such which is not
contained in any strictly larger proper ideal of A.
14
1.5.5 Lemma. Let A be a unital Banach algebra. If I is an ideal of A, then
I is a proper ideal if and only if I Inv A = .
Proof. We have 1 Inv A, so if I Inv A = then 1 , I, so I ,= A and I is
a proper ideal. Conversely, if I Inv A ,= , let b I Inv A. If a A then
a = (ab
1
)b I, since I is an ideal and b I. So I = A.
1.5.6 Theorem. Let A be a unital Banach algebra.
(i). If I is a proper ideal of A then the closure I is also a proper ideal of A.
(ii). Any maximal ideal of A is closed.
Proof. (i) The closure of a vector subspace of A is again a vector subspace.
If a A and x
n
is a sequence in I converging to x I then ax
n
ax and
x
n
a xa as n . Since each ax
n
and x
n
a is in I, this shows that ax and
xa are in I, which is therefore an ideal of A.
Since I is a proper ideal we have I Inv A = by Lemma 1.5.5. Since
Inv A is open by Corollary 1.2.8, this shows that I Inv A = so I ,= A.
(ii) Let M be a maximal ideal. Since M M and M is a proper ideal
by (i), we must have M = M, so M is closed.
1.5.7 Remarks. (i). Since we know a few results about ideals of rings,
we would like to apply these to ideals of Banach algebras. Any unital
Banach algebra A may be viewed as a unital ring R by ignoring the
norm and scalar multiplication. However, there is a dierence in the
denitions: ideals of R are not required to be linear subspaces (since
R has no linear structure) whereas ideals of A are. However, the two
denitions turn out to be equivalent if A is unital. Indeed, an ideal
of the Banach algebra A is clearly an ideal of the ring R. Conversely,
if I is an ideal of the ring R then since 1 R for C we have
x = 1 x I for all x I, so I is a vector subspace of A with the
ideal property. So I is an ideal of A.
(ii). By [FA2.16], any proper ideal of a unital Banach algebra A is contained
in a maximal ideal of A.
1.5.8 Denition. Let A and B be Banach algebras. A homomorphism from
A to B is a linear map : A B which is multiplicative in the sense that
(ab) = (a)(b) for all a, b A.
The kernel of such a homomorphism is the set
ker = a A: (a) = 0.
15
If A and B are unital Banach algebras, we say that a homomorphism
: A B is unital if (1
A
) = 1
B
.
A bijective homomorphism : A B is an isomorphism. If such an
isomorphism exists then the Banach algebras A and B are isomorphic. It is
easy to see that if is an isomorphism then so is
1
.
1.5.9 Remark. If : A B is a non-zero homomorphism of Banach al-
gebras then ker is an ideal of A, which is proper unless = 0. If is
continuous then ker is a closed ideal of A.
1.5.10 Remark. We usually say that two objects are isomorphic if they
have the same structure; that is, if they are the same up to relabelling.
However, if two Banach algebras A and B are isomorphic then this tells us
that they have the same structure as algebras, but not necessarily as Banach
algebras, since the norms may not be related.
The strongest notion of the same Banach algebra up to relabelling is
isometric isomorphism. Two Banach algebras A and B are isometrically
isomorphic if there is an isomorphism : A B which is also an isometry,
meaning that |(a)| = |a| for all a A (compare with [FA1.3.8]).
1.5.11 Examples. (i). If A is a non-unital Banach algebra then the map
: A

A, a (a, 0) from Denition 1.3.20 is an isometric homomor-
phism. Hence (A) is a Banach subalgebra of

A which is isometrically
isomorphic to A.
(ii). The map A(D) A(T), f f[
T
from Example 1.3.14 is a unital
isometric isomorphism.
1.5.12 Proposition. Let A and B be unital Banach algebras and let : A
B be a unital homomorphism.
(i). (Inv A) Inv B, and (a)
1
= (a
1
) for a Inv A.
(ii). For all a A we have
A
(a)
B
((a)).
(iii). If is an isomorphism then
A
(a) =
B
((a)) for all a A.
Proof. (i) If a Inv Athen (a)(a
1
) = (aa
1
) = (1) = 1 and (a
1
)(a) =
(a
1
a) = (1) = 1, so (a) is invertible in B, with inverse (a
1
).
(ii) If
B
((a)) then (a) = ( a) , Inv B so a , Inv A
by (i). Hence
A
(a).
(iii) Since
1
is a homomorphism, by (ii) we have

A
(a) =
A
(
1
((a)))
B
((a))
A
(a),
and we have equality.
16
2 A topological interlude
2.1 Topological spaces
Recall that a topological space is a set X with a topology: a collection T of
subsets of X, known as open sets, such that and X are open, and nite
intersections and arbitrary unions of open sets are open. We call a set F X
closed if its complement X F is open.
An open cover C of X is a collection of open subsets of X whose union
is X. A nite subcover of C is a nite subcollection whose union still con-
tains X. To say that X is compact means that every open cover of X has a
nite subcover. Similarly, if Z X then an open cover C of Z is a collection
of open subsets whose union contains Z. We say that Z is compact if every
open cover of Z has a nite subcover. If X is compact, then it is easy to
show that any closed subset of X is also compact.
A topological space Y is Hausdor if, for any two distinct points y
1
, y
2
in Y , there are disjoint open sets G
1
, G
2
Y with y
1
G
1
and y
2
G
2
. It
is not hard to show that any compact subset of a Hausdor space is closed.
If X and Y are topological spaces then a map : X Y is continuous if,
for all open sets G Y , the set
1
(G) is open in X. Taking complements,
we see that is continuous if and only if, for all closed K Y , the set
1
(K)
is closed in X. If Z is a compact subset of X and : X Y is continuous,
then (Z) is compact.
A homeomorphism from X to Y is a bijection X Y which is continuous
and has a continuous inverse. If there is a homeomorphism from X to Y ,
we say that X and Y are homeomorphic. If X and Y are homeomorphic
topological spaces then all of their topological properties are identical. In
particular, X is compact if and only if Y is compact.
Recall that if X is a topological space and Z X, then the subspace
topology on Z is dened by declaring the open sets of Z to be the sets GZ
for G an open set of X. Then Z is a compact subset of X if and only if Z is
a compact topological space (in the subspace topology). Also, it is easy to
see that a subspace of a Hausdor space is Hausdor.
2.1.1 Lemma. Let X and Y be topological spaces, and suppose that X is
compact and Y is Hausdor.
(i). If : X Y is a continuous bijection, then is a homeomorphism
onto Y . In particular, Y is compact.
(ii). If : X Y is a continuous injection, then (X) (with the subspace
topology from Y ) is homeomorphic to X. In particular, (X) is a
compact subset of Y .
17
Proof. (i) Let K be a closed subset of X. Since X is compact, K is compact.
Since is continuous, (K) is compact. A compact subset of a Hausdor
space is closed, so (K) is closed. Hence
1
is continuous, which shows that
is a homeomorphism. So Y is homeomorphic to the compact space X; so
Y is compact.
(ii) The subspace (X) of the Hausdor space Y is Hausdor. Let

: X (X), x (x), which is a continuous bijection. By (i),

is a
homeomorphism and (X) is compact.
2.2 Subbases and weak topologies
2.2.1 Denition. If T
1
and T
2
are two topologies on a set X then we say
that T
1
is weaker than T
2
if T
1
T
2
.
[We might also say that T
1
is smaller, or coarser than T
2
].
2.2.2 Denition. Let (X, T ) be a topological space. We say that a collection
of open sets o T is a subbase for T if T is the weakest topology containing
o. If the topology T is understood, we will also say that o is a subbase for
the topological space X.
2.2.3 Remark. Suppose that T is a topology on X and o T . It is not
hard to see that the collection of unions of nite intersections of sets in o
forms a topology on X which is no larger than T . [The empty set is the
union of zero sets, and X is the intersection of zero sets, so and X are
in this collection.] From this, it follows that that following conditions are
equivalent:
(i). o is a subbase for T ;
(ii). every set in T is a union of nite intersections of sets in o;
(iii). a set G X is in T if and only if for every x G, there exist nitely
many sets S
1
, S
2
, . . . , S
n
o such that
x S
1
S
2
S
n
G.
By the equivalence of (i) and (ii), if X, Y are topological spaces and o is
a subbase for X, then a map f : Y X is continuous if and only if f
1
(S)
is open for all S o.
18
2.2.4 Proposition. Let X be a set, let I be an index set and suppose that
X
i
is a topological space and f
i
: X X
i
for each i I. The collection
o = f
1
i
(G): i I, G is an open subset of X
i

is a subbase for a topology on X, and this is the weakest topology such that
f
i
is continuous for all i I.
Proof. Let T be the collection of all unions of nite intersections of sets
from o. Then T is a topology on X and o is a subbase for T by the previous
remark. If T

is any topology on X such that f

i
: X X
i
is continuous
for all i I then by the denition of continuity, f
1
i
(G) T

for all open

subsets G X
i
, so o T

. Since T is the weakest topology containing o

this shows that T T

, so T is the weakest topology such that each f

i
is
continuous.
This allows us to introduce the following terminology.
2.2.5 Denition. If X is a set, X
i
is a topological space and f
i
: X X
i
for i I then the weakest topology on X such that f
i
is continuous for each
i I is called the weak topology induced by the family f
i
: i I.
2.2.6 Proposition. Suppose that X is a topological space with the weak
topology induced by a family of maps f
i
: i I where f
i
: X X
i
and X
i
is a topological space for each i I.
If Y is a topological space then a map g : Y X is continuous if and
only if f
i
g : Y X
i
is continuous for all i I.
Proof. The sets f
1
i
(G) for i I and G an open subset of X
i
form a subbase
for X, by Proposition 2.2.4. Hence, by Remark 2.2.3,
g is continuous g
1
(f
1
i
(G)) is open for all i I and open G X
i
(f
i
g)
1
(G) is open for all i I and open G X
i
f
i
g is continuous for all i I.
2.2.7 Lemma. Suppose that X is a topological space with the weak topology
induced by a family of mappings f
i
: X X
i

iI
. If Y X then the weak
topology induced by the family f
i
[
Y
: Y X
i

iI
is the subspace topology
on Y .
Proof. Let g
i
= f
i
[
Y
for i I. Observe that g
1
i
(G) = f
1
i
(G) Y for G an
open subset of X
i
. It is not hard to check that the collection of all sets of
this form is a subbase for both the weak topology induced by g
i

iI
and for
the subspace topology on Y . Hence these topologies are equal.
19
2.3 The product topology and Tychonos theorem
If o is a collection of open subsets of X, let us say that an open cover is an
o-cover if every set in the cover is in o.
2.3.1 Theorem (Alexanders subbase lemma). Let X be a topological space
with a subbase o. If every o-cover of X has a nite subcover, then X is
compact.
Proof. Suppose that the hypothesis holds but that X is not compact. Then
there is an open cover with no nite subcover; ordering such covers by in-
clusion we can apply Zorns lemma [FA2.15] to nd an open cover C of X
without a nite subcover that is maximal among such covers.
Note that C o cannot cover X by hypothesis, so there is some x X
which does not lie in any set in the collection C o. On the other hand, C
does cover X so there is some G C o with x G. Since G is open and
o is a subbase, by Remark 2.2.3 we have
x S
1
S
n
G
for some S
1
, . . . , S
n
o. For i = 1, . . . , n we have x S
i
so S
i
, C. By the
maximality of C, there is a nite subcover C
i
of CS
i
. Let D
i
= C
i
S
i
.
Then D = D
1
D
n
covers X (S
1
S
n
). So D G is a nite
subcover of C, which is a contradiction. So X must be compact.
2.3.2 Denition. Let X
i
: i I be an indexed collection of sets. Just as
in [FA2.4], we dene the (Cartesian) product of this collection to be the set

iI
X
i
= f : I
_
iI
X
i
: f(i) X
i
for all i I.
If each X
i
is a topological space then the product topology on X =

iI
X
i
is the weak topology induced by the family
i
: i I where the map
i
is
the evaluation at i map

i
: X X
i
, f f(i).
2.3.3 Remark. If f

iI
X
i
then it is often useful to think of f as the
I-tuple (f(i))
iI
. In this notation, we have

iI
X
i
= (x
i
)
iI
: x
i
X
i
for all i I
and
i
: (x
i
)
iI
x
i
is the projection onto the ith coordinate.
20
2.3.4 Theorem (Tychonos theorem).
The product of a collection of compact topological spaces is compact.
Proof. Let X
i
be a compact topological space for i I and let X =

iI
X
i
with the product topology. Consider the collection
o =
1
i
(G): i I, G is an open subset of X
i
.
By Proposition 2.2.4, o is a subbase for the topology on X.
Let C be an o-cover of X. For i I, let C
i
= G X
i
:
1
i
(G) C,
which is a collection of open subsets of X
i
.
We claim that there is i I such that C
i
is a cover of X
i
. Otherwise,
for every i I there is some x
i
X
i
not covered by C
i
. Consider the map
f X dened by f(i) = x
i
. By construction, f does not lie in
1
i
(G) for
any i I and G C
i
. However, since C o, every set in C is of this form,
so C cannot cover f. This contradiction establishes the claim.
So we can choose i I so that C
i
covers X
i
. Since X
i
is compact, there
is nite subcover D
i
of C
i
. But then
1
i
(G): G D
i
covers X, and this
is a nite subcover of C.
This shows that every o-cover of X has a nite subcover. By Theo-
rem 2.3.1, X is compact.
2.4 The weak* topology
Let X be a Banach space. Recall from [FA3.2] that the dual space X

of X
is the Banach space of continuous linear functionals : X C, with the
norm || = sup
x1
[(x)[.
2.4.1 Denition. For x X, let J
x
: X

C, (x). The weak*

topology on X

is the weak topology induced by the family J

x
: x X.
2.4.2 Remarks. (i). For x X, the map J
x
is simply the canonical image
of x in X

. In particular, each J
x
is continuous when X

is equipped
with the usual topology from its norm, so the weak* topology is weaker
(that is, no stronger) than the norm topology on X

. In fact, the weak*

topology is generally strictly weaker than the norm topology.
(ii). The sets X

: [(x) (x)[ < for X

, > 0 and x X
form a subbase for the weak* topology.
(iii). By (ii), it is easy to see that X

with the weak* topology is a Hausdor

topological space.
21
2.4.3 Theorem (The Banach-Alaoglu theorem). Let X be a Banach space.
The closed unit ball of X

is compact in the weak* topology.

Proof. For x X let D
x
= C: [[ |x|. Since D
x
is closed and
bounded, it is a compact topological subspace of C. Let D =

xX
D
x
with the product topology. By Tychonos theorem 2.3.4, D is a compact
topological space.
Let X

1
= X

: || 1 denote the closed unit ball of X

, with the
subspace topology that it inherits from the weak* topology on X

. We must
show that X

1
is compact.
If is any linear map X C, then X

1
if and only if [(x)[ |x|
for all x X, i.e. (x) D
x
for all x X. Thus X

1
D. Moreover, if
x X and X

1
then
J
x
() = (x) =
x
(),
so J
x
[
X

1
=
x
[
X

1
. By Lemma 2.2.7, the topology on X

1
is equal to the
subspace topology when we view it as a subspace of D.
Since D is compact, it suces to show that X

1
is a closed subset of D.
Now
X

1
= D: is linear =

,C,
x,yX
D:
x+y
() =
x
() +
y
().
By the denition of the product topology on D, the maps
x
: D D
x
are
continuous for each x X. Linear combinations of continuous functions are
continuous, so for x, y, z X and , C, the function =
z

y
is
continuous D C. Hence
1
(0) is closed. Each set in the above intersection
is of this form, so X

1
is closed.
22
3 Unital abelian Banach algebras
3.1 Characters and maximal ideals
Let A be a unital abelian Banach algebra.
3.1.1 Denition. A character on A is a non-zero homomorphism A C;
that is, a non-zero linear map : A C which satises (ab) = (a)(b) for
a, b A. We write (A) for the set of characters on A.
3.1.2 Example. Let A = C(X) where X is a compact topological space.
For each x X, the map
x
: A C, f f(x) is a character on A.
3.1.3 Remark. This denition makes sense even if A is not abelian. How-
ever, (A) is often not very interesting in that case.
For example, if A = M
n
(C) and n > 1 then (A) = . Indeed, it is not
hard to show that A is spanned by ab ba: a, b A. If : A C is a
homomorphism, then (ab ba) = (a)(b) (b)(a) = 0, so = 0 by
linearity; hence (A) = .
3.1.4 Lemma. If (A) then is continuous. More precisely,
|| = (1) = 1.
In particular, (A) is a subset of the closed unit ball of A

.
Proof. Observe that (1) = (1
2
) = (1)
2
, so (1) 0, 1. If (1) = 0 then
(a) = (a1) = (a)(1) = 0 for any a A, so = 0. But (A) so
,= 0, which is a contradiction. So (1) = 1.
We have (Inv A) Inv C = C 0 by 1.5.12(i). For any a A we have
((a)1a) = (a)(1) (a) = 0, so (a)1a , Inv A. Hence (a) (a),
so [(a)[ |a| by Theorem 1.3.5 and so || 1. Since [(1)[ = 1 we
conclude that || = 1.
Any (A) is a linear map A C by denition, and we have shown
that it continuous with norm 1. Hence (A) is contained in the closed unit
ball of A

.
3.1.5 Denition. The Gelfand topology on (A) is the subspace topology
obtained from the weak* topology on A

.
We will always equip (A) with this topology.
23
3.1.6 Theorem. (A) is a compact Hausdor space.
Proof. We observed in Remark 2.4.2(iii) that the weak* topology is Haus-
dor, so (A) is a Hausdor space. By Lemma 3.1.4, (A) is contained in the
unit ball of A

, which is compact in the weak* topology by Theorem 2.4.3.

A closed subset of a compact set is compact, so it suces to show that (A)
is weak* closed in A

. But
(A) = A

: (1) = 1, (ab) = (a)(b) for a, b A

= A

: (1) = 1

a,bA
A

: (ab) (a)(b) = 0
= J
1
1
(1)

a,bA
(J
ab
J
a
J
b
)
1
(0).
Each evaluation functional J
a
: A

C, (a) is weak* continuous, so

the maps J
ab
J
a
J
b
are also weak* continuous. Hence the sets in this
intersection are all weak* closed, so (A) is weak* closed.
3.1.7 Lemma. Let A be a unital abelian Banach algebra.
(i). If (A) then ker is a maximal ideal of A.
(ii). If M is a maximal ideal of A, then the map C A/M, 1 + M
is an isometric isomorphism.
Proof. (i) Let (A). Since is a non-zero homomorphism, its kernel
I = ker is a proper ideal of A. Suppose that J is an ideal of A with I J
and let a J I. Then (a) ,= 0, so b = (a)
1
a J and (b) = 1. Since
(1) = 1 by Lemma 3.1.4, we have 1 b I, so 1 = b + 1 b J. By
Lemma 1.5.5, J = A. This shows that I is not contained in any strictly
larger proper ideal of A, so I is a maximal ideal of A.
(ii) Let M be a maximal ideal of A. By Theorems 1.5.6(ii) and 1.5.3,
A/M is unital Banach algebra with unit 1 + M. If a + M is a non-zero
element of A/M then a A M. Let I = ab + m: m M, b A. Since
A is abelian and M is an ideal, it is easy to see that I is an ideal of A, and
M I. Since M is a maximal ideal, I = A. So 1 I, and ab + m = 1 for
some b A and m M. Now
(a + M)(b + M) = ab + M = ab + m + M = 1 + M,
which is the unit of A/M. Hence b +M = (a+M)
1
and a+M is invertible
in A/M. By the Gelfand-Mazur theorem 1.3.6, A/M = C1
A/M
= C(1 +M).
It is very easy to check that the map C A/M, 1 + M is an
isometric homomorphism, and we have just shown that it is surjective. Hence
it is an isometric isomorphism.
24
3.1.8 Theorem. Let A be a unital abelian Banach algebra. The mapping
ker
is a bijection from (A) onto the set of maximal ideals of A.
Proof. If (A) then ker is a maximal ideal of A, by Lemma 3.1.7(i).
Hence the mapping is well-dened.
The mapping ker is injective, since if
1
and
2
are in (A) with
ker
1
= ker
2
, then for any a A we have a
2
(a)1 ker
2
= ker
1
so

1
(a
2
(a)1) = 0, hence
1
(a) =
2
(a); so
1
=
2
.
We now show that the mapping is surjective. Let M be a maximal ideal
of A, and let q : A A/M, a a +M be the corresponding quotient map.
Observe that q is a homomorphism and ker q = M. By Lemma 3.1.7(ii), the
map : C A/M, 1+M is an isomorphism. Let =
1
q : A C.
Since is the composition of two homomorphisms, it is a homomorphism,
and (1) =
1
(q(1)) =
1
(1 + M) = 1, so ,= 0. Hence (A). Since
is an isomorphism, we have ker = ker q = M.
This shows that ker is a bijection from (A) onto the set of all
maximal ideals of A.
3.1.9 Examples. (i). Let X be a compact Hausdor space. For x X,
the map
x
: C(X) C, f f(x) is a nonzero homomorphism, so

x
: x X (C(X)). We claim that we have equality.
For x X, let M
x
= ker
x
= f C(X): f(x) = 0, which is a
maximal ideal of C(X) by Theorem 3.1.8. Let I be an ideal of C(X).
If I , M
x
for every x X, then for each x X, there is f
x
I with
f
x
,= 0. Since I is an ideal, g
x
= [f
x
[
2
= f
x
f
x
I, and since g
x
is
continuous and non-negative with g
x
(x) > 0, there is an open set U
x
with x U
x
and g
x
(y) > 0 for all y U
x
. As x varies over X, the open
sets U
x
cover X. Since X is compact, there is n 1 and x
1
, . . . , x
n
X
such that U
x
1
, . . . , U
x
n
cover X. Let g = g
x
1
+ +g
x
n
. Then g I and
g(x) > 0 for all x X, so g is invertible in C(X). Hence I = C(X).
This shows that every proper ideal I of C(X) is contained in M
x
for
some x X. Let (C(X)). Since ker is a maximal (proper) ideal,
we must have ker = M
x
for some x X, so =
x
by Theorem 3.1.8.
Consider the map : X (C(X)), x
x
. We have just shown that
this is surjective. Since X is compact and Hausdor, C(X) separates
the points of X by Urysohns lemma. Hence if
x
=
y
then f(x) = f(y)
for all f C(X), so x = y. Hence is a bijection.
25
We claim that is a homeomorphism. Indeed, is continuous since for
f C(X) and x X we have
J
f
((x)) = J
f
(
x
) =
x
(f) = f(x),
so J
f
(or, more precisely, J
f
[
(C(X))
) is continuous X C for
every f C(X). By Proposition 2.2.6, is continuous. Since X is
compact and (C(X)) is Hausdor, Lemma 2.1.1 shows that is a
homeomorphism.
(ii). Recall that A(D) denotes the disc algebra. If w D then
w
: A(D)
C, f f(w) is a character on A(D).
Again, we claim that every character arises in this way. To see this,
consider the function z A(D) dened by z(w) = w, w D. If
(A(D)) then (1) = 1 and [(z)[ |z| = 1, so (z) D.
It is not hard to show that the polynomials D C form a dense
unital subalgebra of A(D). If p: D C is a polynomial then p =

0
1 +
1
z + +
n
z
n
for constants
i
, so (p) =
0
+
1
(z) + +

n
(z)
n
= p((z)). Since the polynomials are dense in A(D), this
shows that (f) = f((z)) for all f A(D), so =
(z)
. Hence
(A(D)) =
w
: w D. Just as in (i), it is easy to see that the map
w
w
is a homeomorphism D (A(D)).
(iii). Recall the abelian Banach algebra
1
(Z) with product from Exam-
ple 1.1.3(v). For n Z, let e
n
= (
mn
)
mZ
. Then e
n

1
(Z), and the
linear span of e
n
: n Z is dense in
1
(Z). Moreover, it is easy to
check that e
n
e
m
= e
n+m
for n, m Z. In particular, e
0
e
m
= e
m
,
hence e
0
is the unit for
1
(Z).
If (
1
(Z)) then (e
0
) = 1. Moreover, e
n
e
n
= e
0
so e
n
= (e
n
)
1
and [(e
n
)[ |e
n
| = 1 for each n Z, so
1 [(e
n
)[
1
= [((e
n
)
1
)[ = [(e
n
)[ 1.
So we have equality. In particular, [(e
1
)[ = 1, so (e
1
) T. Since
e
n
= e
n
1
, we have (e
n
) = (e
1
)
n
. Hence if x
1
(Z) then
(x) =
_

nZ
x
n
e
n
_
=

nZ
x
n
(e
1
)
n
.
So is determined by the complex number (e
1
) T. Conversely, given
any z T, there is a character
z
(
1
(Z)) with
z
(e
1
) = z. Indeed,
let A
0
= spane
n
: n Z, which is a dense subalgebra of
1
(Z). Let
26

0
: A
0
C be the unique linear map such that
0
(e
n
) = z
n
for all
n Z. If x, y A
0
, then

0
(x y) =
0
_

m,nZ
x
m
y
nm
e
n
_
=

m,nZ
x
m
y
nm
z
n
=

m,nZ
x
m
z
m
y
nm
z
nm
=
0
(x)
0
(y),
so
0
is a homomorphism, and
[
0
(x)[ =

nZ
x
n
z
n

nZ
[x
n
[ = |x|,
so
0
is continuous. Hence
0
extends to a continuous linear homomor-
phism
z
:
1
(Z) C, which is a character on
1
(Z). Clearly,
z
(e
1
) = z.
This shows that the map : T (
1
(Z)), z
z
is a bijection. We
claim that is a homeomorphism. Indeed, is continuous since for
x
1
(Z) and z T we have
J
x
((z)) = J
x
(
z
) =
z
(x) =

nZ
x
n
z
n
.
Since x
1
(Z), this series converges absolutely (for z T). The partial
sums of the series are continuous functions T C, so z J
x
((z))
is continuous T C. By Proposition 2.2.6, is continuous. Since T
is compact and (
1
(Z)) is Hausdor, Lemma 2.1.1 shows that is a
homeomorphism.
The next lemma is purely algebraic.
3.1.10 Lemma. Let A be a unital abelian Banach algebra and let a A.
The following are equivalent:
(i). a , Inv A;
(ii). a I for some proper ideal I of A;
(iii). a M for some maximal ideal M of A.
Proof. (i) = (ii): If a , Inv A, consider the set I = ab: b A. Since A
is abelian, this is an ideal of A, and since A is unital we have a = a1 I. If
1 I then ab = 1 for some b A, so a Inv A, a contradiction. So 1 , I
and I is a proper ideal.
(ii) = (iii): Suppose that a I where I is a proper ideal of A. By
Remark 1.5.7(ii), I M for some maximal ideal M of A; so a M.
(iii) = (i): If M is a maximal ideal of A then M is a proper ideal, so
M Inv A = by Lemma 1.5.5. Hence a , Inv A for every a M.
27
3.1.11 Corollary. Let A be a unital abelian Banach algebra and let a A.
(i). a Inv A if and only if (a) ,= 0 for all (A).
(ii). (a) = (a): (A).
(iii). r(a) = sup
(A)
[(a)[.
Proof. (i) We have:
a Inv A a , M for all maximal ideals M of A, by Lemma 3.1.10
a , ker for all (A), by Theorem 3.1.8
(a) ,= 0 for all (A).
(ii) This follows from the equivalences:
(a) a , Inv A
( a) = 0 for some (A), by (i)
= (a) for some (A), since () = by Lemma 3.1.4.
(iii) follows immediately from (ii) and the denition of r(a).
3.2 The Gelfand representation
3.2.1 Denition. Let A be a unital abelian Banach algebra. For a A, the
Gelfand transform of a is the mapping
a: (A) C, (a).
In other words, a = J
a
[
(A)
.
3.2.2 Examples. (i). Let X be a compact Hausdor space. We have seen
that the map X (C(X)), x
x
is a homeomorphism. If f
C(X) then

f : (C(X)) C,
x

x
(f) = f(x).
This means that, if we identify (C(X)) with X by pretending that
x =
x
, then

f = f.
(ii). We have seen that (
1
(Z)) can be identied with T, by pretending that
z T is the same as the character
z
:
1
(Z) C, x

nZ
x
n
z
n
.
Hence for x
1
(Z), we have
x: (
1
(Z)) C,
z

z
(x) =

nZ
x
n
z
n
.
28
If we write z = e
i
and x
n
= x(n) then this takes the form
x(e
i
) =

nZ
x(n)e
in
,
so x maybe viewed as the inverse Fourier transform of x.
3.2.3 Theorem. Let A be a unital abelian Banach algebra. For each a A,
the Gelfand transform a is in C((A)). Moreover, the mapping
: A C((A)), a a
is a unital, norm-decreasing (and hence continuous) homomorphism, and for
each a A we have

A
(a) =
C((A))
(a) = a(): (A) and r(a) = |a|.
Proof. By the denition of the topology on (A), each a is in C((A)). It
is easy to see that is a homomorphism, and it is unital by Lemma 3.1.4.
The identication of
A
(a) with
C((A))
(a) follows from Corollary 3.1.11(ii)
and Example 1.3.2(ii). Now r(a) = |a| |a| by Corollary 3.1.11(iii) and
Remark 1.3.10, so is linear and norm-decreasing, hence continuous.
3.2.4 Denition. If A is a unital abelian Banach algebra then the unital
homomorphism : A C((A)), a a is called the Gelfand representation
of A.
3.2.5 Remark. In general, the Gelfand representation is neither injective
nor surjective. For example, the Gelfand representation of the disc algebra
is the inclusion A(D) C(D) which is not surjective.
29
4 C*-algebras
4.1 Denitions and examples
4.1.1 Denition. Let A be a Banach algebra. An involution on A is a map
A A, a a

such that for all a, b A and C we have:

(i). (a)

= a

and (a + b)

= a

+ b

(conjugate linearity);
(ii). (ab)

= b

; and
(iii). (a

= a.
A C*-algebra is a Banach algebra A equipped with an involution such that
the C*-condition holds:
|a|
2
= |a

a| for all a A.
4.1.2 Remark. We usually write a

instead of (a

.
4.1.3 Examples. (i). The zero Banach algebra 0 is a C*-algebra.
(ii). The complex numbers C form a C*-algebra under the usual Banach
space norm || = [[ and the involution

= .
(iii). If X is a topological space, then BC(X) is a C*-algebra under the
involution f

(x) = f(x). In particular, if X is compact, then BC(X) =

C(X) is a C*-algebra.
(iv). If H is a Hilbert space then B(H) is a C*-algebra under the involution
T T

dened by the property that Tx, y) = x, T

y) for all x, y H
(see [FA]).
(v). If A is a C*-algebra then a C*-subalgebra of A is a Banach subalgebra
B A that is closed under the involution; in other words, B is a closed
linear subspace of A such that whenever a, b A we have ab A and
a

A. Clearly, any C*-subalgebra of a C*-algebra is a C*-algebra.

4.1.4 Proposition. Let A be a C*-algebra.
(i). If A has an identity element 1 then 1

= 1, and if A is non-zero then

|1| = 1.
(ii). If A is unital then a Inv A a

Inv A, and for a Inv A we

have (a

)
1
= (a
1
)

.
(iii). |a

| = |a| for all a A.

30
(iv). The involution A A, a a

is continuous.
(v). (a

) = (a)

= : (a) for all a A.

Proof. (i) We have
1 = (1

= (1

1)

= 1

= 1

1 = 1

.
Hence |1|
2
= |1

1| = |1|; if A ,= 0 then |1| ,= 0 and we can cancel to

obtain |1| = 1.
(ii) If a Inv A then a

(a
1
)

= (a
1
a)

= 1

= 1 and (a
1
)

=
(aa
1
)

= 1

= 1. Hence a

is invertible, with inverse (a

1
)

. Conversely, if
a

Inv A then a

= a Inv A by the same argument.

(iii) This is trivial for a = 0. If a ,= 0 then |a|
2
= |a

a| |a

| |a|,
which on cancelling gives |a| |a

|. Hence |a

| |a

| = |a| |a

|, so
we have equality.
(iv) If a
n
, a A with a
n
a then |a

n
a

| = |(a
n
a)

|
(iii)
= |a
n
a| 0
as n , so the involution is continuous.
(v) Without loss of generality, suppose that A is unital. For any C,
(a

) 1a

(i)
= (1a)

, Inv A
(ii)
1a , Inv A (a).
Hence (a

) = (a)

.
4.1.5 Denition. Let A be a C*-algebra.
An element a A is normal if a commutes with a

.
An element a A is hermitian if a = a

.
An element p A is a projection if p = p

= p
2
.
If A is unital then an element u A is unitary if uu

= u

u = 1 (that is,
if u is invertible and u
1
= u

).
4.1.6 Remark. Clearly, projections are hermitian, and both unitary and
hermitian elements are normal.
4.1.7 Proposition. Let A be a C*-algebra.
(i). If a A then a

a is hermitian.
(ii). If p A is a non-zero projection then |p| = 1.
(iii). Every a A may be written uniquely in the form a = h + ik where h
and k are hermitian elements of A, called the real and imaginary parts
of a.
31
Proof. (i) If h = a

a then h

= (a

a)

= a

= a

a = h, so h is hermitian.
(ii) We have |p|
2
= |p

p| = |p
2
| = |p|. Since |p| ,= 0 we may cancel
to obtain |p| = 1.
(iii) We have a = h + ik where h =
1
2
(a + a

) and k =
1
2i
(a a

), as
is easily veried. If h + ik = h

+ ik

where h, h

, k, k

are hermitian, then

i(k

k) = h h

= (h h

=
_
i(k

k)
_

= i(k

k), hence h = h

and
k = k

, demonstrating uniqueness.
4.1.8 Lemma. If A is a unital C*-algebra and A

with || 1 and
(1) = 1, then (h) R for all hermitian elements h A.
Proof. Suppose that (h) = x + iy where x, y R. Observe that for t R
we have (h + it1) = (h) + it = x + i(y + t), so
x
2
+(y+t)
2
= [(h+it)[
2
|h+it|
2
= |(hit)(h+it)| = |h
2
+t
2
| |h
2
|+t
2
and so x
2
+2yt |h
2
| for all t R. This forces y = 0, so (h) = x R.
4.1.9 Proposition. If A is a unital abelian C*-algebra then (a

) = (a)
for all a A and (A).
Proof. If (A) then || = (1) = 1 by Lemma 3.1.4. If a A then
a = h +ik for some hermitian h, k A by Proposition 4.1.7(iii), so (h) and
(k) are real by Lemma 4.1.8. Hence
(a

) = (h ik) = (h) i(k) = (h) +i(k) = (a).

4.1.10 Corollary. Let A be a unital C*-algebra.
(i). If h A is hermitian then
A
(h) R.
(ii). If u A is unitary then
A
(u) T.
Proof. If a A and a commutes with a

then let C = a, a

. This is a
closed unital abelian subalgebra of A and
A
(a) =
C
(a) = (a): (C)
by Proposition 1.3.19 and Theorem 3.2.3. The commutant of a self-adjoint
subset of A is self-adjoint (that is, it is closed under the involution), so C is
self-adjoint. Hence C is a unital abelian C*-subalgebra of A.
(i) If a = h is hermitian then (a) R for all (C) by Proposi-
tion 4.1.9, which establishes the result.
(ii) If a = u is unitary then by Proposition 4.1.9 we have
[(u)[
2
= (u)

(u) = (u

)(u) = (u

u) = (1) = 1
for every (C), so
A
(u) T.
32
4.1.11 Corollary. Let A be a unital C*-algebra and let B A be a C*-
subalgebra with 1 B. If b B then
B
(b) =
A
(b).
Proof. If b is hermitian then
A
(b) R by Corollary 4.1.10(i). Therefore

A
(b) has no holes, so
B
(b) =
A
(b) by Theorem 1.3.16.
If b is any element of BInv A and if a is the inverse of b in A then b

b is
invertible in A with inverse aa

. Since b

b is hermitian, the rst paragraph

shows that b

b is invertible in B; so aa

B. Hence aa

B and baa

=
aa

b = 1, so b is invertible in B with inverse aa

. Hence b Inv B.
This shows that Inv B = B Inv A, so
B
(b) =
A
(b) for any b B.
4.1.12 Proposition.
If a is a hermitian element of a C*-algebra then |a| = r(a).
Proof. Since a = a

we have |a|
2
= |a

a| = |a
2
|, so for n 1 we have
|a|
2
n
= |a
2
n
| by induction. Hence, by Theorem 1.3.12,
r(a) = lim
n
|a
2
n
|
1/2
n
= |a|.
4.1.13 Denition. If A and B are C*-algebras then a -homomorphism
: A B is a homomorphism of Banach algebras which respects the involu-
tion: (a

) = (a)

for all a A. If A and B are unital and (1) = 1 then we

say that is unital. An invertible -homomorphism is called a -isomorphism.
4.1.14 Proposition. Let A and B be unital C*-algebras and let : A B
be a unital -homomorphism.
(i). (Inv A) Inv B, and (a)
1
= (a
1
) for a Inv A.
(ii). For all a A we have
A
(a)
B
((a)).
(iii). is continuous; in fact, |(a)| |a| for all a A.
(iv). If is a -isomorphism then is an isometry and
A
(a) =
B
((a))
for all a A.
Proof. (i) and (ii) are special cases of Proposition 1.5.12.
(iii) Let a A. We have
|(a)|
2
= |(a)

(a)| = r
B
((a)

(a)) by Proposition 4.1.12

= r
B
((a

a))
r
A
(a

a) by (ii)
= |a

a| by Proposition 4.1.12
= |a|
2
.
33
So |(a)| |a| for all a A. Since is a linear map, this shows that is
continuous (with || 1).
(iv) If is a -isomorphism then is a bijective -homomorphism, from
which it is easy to see that
1
is also a -homomorphism. By (iii) both
and
1
are continuous linear maps with norm no greater than 1, so
|a| = |
1
((a))| |(a)| |a|
for each a A. Hence we have equality throughout and is an isometry. We
showed that
A
(a) =
B
((a)) for all a A in Proposition 1.5.12.
4.2 The Stone-Weierstrass theorem
4.2.1 Lemma. If m N then there is a sequence of polynomials p
1
, p
2
, p
3
, . . .
with real coecients such that 0 p
n
(t) 1 for n 1 and 0 t 1, and
p
n
0 uniformly on [0,
1
2m
] and p
n
1 uniformly on [
2
m
, 1].
Proof. A calculus exercise shows that if N N then
1 (1 x)
N
Nx and (1 x)
N

1
Nx
for each x (0, 1).
Let p
n
(t) = 1 (1 t
n
)
m
n
. We have 0 p
n
(t) 1 for t [0, 1], and
sup
t[0,1/2m]
[p
n
(t)[ = sup
t[0,1/2m]
1 (1 t
n
)
m
n
sup
t[0,1/2m]
(mt)
n
2
n
0 as n ,
and similarly,
sup
t[2/m,1]
[1p
n
(t)[ = sup
t[2/m,1]
(1t
n
)
m
n
sup
t[2/m,1]
1
(mt)
n
2
n
0 as n .
Let X be a compact Hausdor space, and let C(X, R) denote the set of
continuous functions X R. This is a real Banach space under the uniform
norm [FA1.7.2], and it is easy to check that the pointwise product turns
it into a real Banach algebra. The basic denitions we made for complex
Banach algebras carry over simply by changing C to R, so we may talk of
subalgebras and unital subalgebras of C(X, R).
If A C(X, R) (or A C(X)), we say that A separates the points of X
if for every pair of distinct points x, y X, there is some f A such that
f(x) ,= f(y).
34
4.2.2 Theorem (The real-valued Stone-Weierstrass theorem). Let X be a
compact Hausdor topological space and let A C(X, R). If A is a unital
subalgebra of C(X, R) which separates the points of X, then A is uniformly
dense in C(X, R).
Proof. We will write statements such as f 1 to mean f(x) 1 for all
x X, and f 1 on K to mean that f(x) 1 for all x K. We split
the proof into several steps.
(i). If x, y X with x ,= y then there is a function f A with f 0 such
that f(x) = 0 and f(y) > 0.
Since A separates the points of X, there is a function g A with
g(x) ,= g(y). Let f = (g g(x)1)
2
. Since A is a unital algebra, f A,
and clearly f 0. Moreover, f(x) = 0 and f(y) > 0.
(ii). If L is a compact subset of X and x X L then there is some f A
with 0 f 1 such that f(x) = 0 and f(y) > 0 for all y L.
By (i), for each y L there is a function f
y
A such that f
y
0,
f
y
(x) = 0 and f
y
(y) > 0. Since f
y
is continuous, there is an open set
U
y
containing y such that f
y
> 0 on U
y
. As y varies over L, the sets
U
y
cover L. Since L is compact, there are y
1
, . . . , y
k
L such that
U
y
1
, . . . , U
y
k
cover L. Let f
0
= f
y
1
+ + f
y
k
and let f = f
0
/|f
0
|.
Then f A, 0 f 1, f(x) = 0 and f(y) > 0 for y L.
(iii). If K, L are disjoint compact subsets of X then there is a function f A
with 0 f 1 such that f
1
4
on K and f
3
4
on L.
Let x K. By (ii), there is a function g
x
A such that 0 g
x
1,
g
x
(x) = 0, and g
x
> 0 on L. Since L is compact, g
x
(L) is compact,
so is contained in a set of the form [s, t] for some s > 0. Hence there
is m N (depending on x) such that g
x

2
m
on L. For x K, let
U
x
= z X: g
x
(z) <
1
2m
. Since x U
x
, these open sets cover K,
so by compactness there are x
1
, . . . , x
k
K such that U
x
1
, . . . , U
x
k
cover K.
By Lemma 4.2.1, for i = 1, . . . , k there is a polynomial p (depending
on i) such that if we write f
i
= p(g
x
i
), then 0 f
i
1, f
i

1
4
on U
x
i
and f
i
(
3
4
)
1/k
on L. Since p is a polynomial and g
x
i
A, we have
f
i
A.
Now let f = f
1
f
2
. . . f
k
. Then f A and f has the desired properties.
(iv). If f C(X, R) then there is g A such that |f g|
3
4
|f|.
We may assume (by considering f/|f|) that |f| = 1. Let h A be
the function obtained by applying (iii) to K = x X: f(x)
1
4

35
and L = x X: f(x)
1
4
, and let g = h
1
2
. Since
1
4
g
1
2
on L, we have [f g[
3
4
on L; similarly, [f g[
3
4
on K. Since

1
2
g
1
2
and
1
4
f
1
4
on X (K L) we have [f g[
3
4
on X (K L). Hence |f g|
3
4
.
(v). A is uniformly dense in C(X, R).
Let f C(X, R). By (iv), there is g
1
A with |f g
1
|
3
4
|f|.
Applying (iv) to the function f g
1
, we obtain a function g
2
A with
|f g
1
g
2
|
3
4
|f g
1
| (
3
4
)
2
|f|. Continuing in this manner, we
obtain functions g
1
, g
2
, g
3
, . . . A such that
|f (g
1
+ + g
n
)| (
3
4
)
n
|f| 0 as n .
Since g
1
+ + g
n
A, this shows that f is in the uniform closure
of A.
4.2.3 Theorem (The complex Stone-Weierstrass theorem). Let X be a com-
pact Hausdor topological space and let A C(X). If A is a unital -
subalgebra of C(X) which separates the points of X, then A is uniformly
dense in C(X).
Proof. For f C(X), let f
1
(x) = Re(f(x)) and f
2
(x) = Im(f(x)). Then
f
1
, f
2
C(X, R). Moreover, f
1
=
1
2
(f + f

) and f
2
=
1
2i
(f f

) are in A
since A is a complex vector space which is closed under the -operation.
Consider the set A
1
= f
1
: f A. Since A separates the points of X,
if x, y X with x ,= y then there is f A such that f(x) ,= f(y). Hence
either f
1
(x) ,= f
1
(y) or f
2
(x) ,= f
2
(y). If f
1
(x) ,= f
1
(y) then f
1
is a function
in A
1
separating x and y, and if f
2
(x) ,= f
2
(y) then g = if = f
2
if
1
A
and f
2
= g
1
is a function in A
1
separating x and y.
This shows that A
1
separates the points of X, and it is easy to see that A
1
is a unital subalgebra of C(X, R). By the real-valued Stone-Weierstrass the-
orem, A
1
is uniformly dense in C(X, R).
If f C(X) then f = f
1
+ if
2
. Given > 0 there is a function g
1
A
1
with |f
1
g
1
|

< /2 and a function g

2
A
1
with |f
2
g
2
|

< /2. Hence

g = g
1
+ ig
2
A and |f g|

|f
1
g
1
|

+ |i(f
2
g
2
)|

< , so A is
uniformly dense in C(X).
4.3 Abelian C*-algebras and the continuous functional
calculus
We now apply Gelfands theory of abelian unital Banach algebras (3) to
show that, up to isometric -isomorphism, every unital abelian C*-algebra is
of the form C(X) for some compact Hausdor space X.
36
4.3.1 Theorem (The Gelfand-Naimark theorem). If A is a unital abelian
C*-algebra then the Gelfand representation of A,
: A C((A)), a a
is a unital isometric -isomorphism of A onto C((A)).
Proof. Theorem 3.2.3 tells us that is a unital, norm-decreasing homomor-
phism with |(a)| = r(a). Moreover, for a A and (A) we have

() = (a

) = (a) = a() = (a)

()
by Proposition 4.1.9, so (a

) = (a)

and is a -homomorphism. Now a

a
is hermitian by Proposition 4.1.7(i), so by Proposition 4.1.12,
|(a)|
2
= |(a)

(a)| = |(a

a)| = r(a

a) = |a

a| = |a|
2
.
Hence is an isometry.
It remains to show that is surjective, for which we appeal to the Stone-
Weierstrass theorem. Recall that (A) is a compact Hausdor space by
Theorem 3.1.6. Since is a unital -homomorphism, its image (A) is a
unital -subalgebra of C((A)). If
1
,
2
(A) with
1
,=
2
then there is
a A with
1
(a) ,=
2
(a), hence a(
1
) ,= a(
2
) and so (A) separates the
points of (A). By the Stone-Weierstrass theorem 4.2.3, (A) is dense in
C((A)). Since is a linear isometry its range is closed. Hence
(A) = (A) = C((A)).
4.3.2 Corollary. If A is a unital abelian C*-algebra and a, b A with (a) =
(b) for all (A), then a = b.
Proof. a =

b, so a = b by the Gelfand-Naimark theorem.

4.3.3 Denition. If S is a subset of a C*-algebra A, then we write C

(S)
for the smallest C*-subalgebra of A containing S. In particular, if A is a
unital C*-algebra and a A then we will write C

(1, a) = C

(1, a).
4.3.4 Lemma. Let A be a C*-algebra and let S A.
(i). The linear span of the elements of the form a
n
1
1
a
n
2
2
. . . a
n
k
k
for k 1,
n
1
, . . . , n
k
1 and a
i
S or a

i
S for 1 i k is dense in C

(S).
(ii). Suppose that A is unital and that B is another unital C*-algebra. If
A = C

(S) and
1
,
2
are two unital -homomorphisms A B such
that
1
(a) =
2
(a) for all a S, then
1
=
2
.
37
Proof. (i) Clearly, any C*-algebra containing S must also contain every ele-
ment of the given form, hence C

(S) contains the closed linear span of such

elements. Conversely, it is easy to see that the closed linear span of these
elements forms a C*-algebra containing S, so this C*-algebra is equal to
C

(S).
(ii) Since
1
and
2
are unital -homomorphisms, they are continuous by
Proposition 4.1.14(iii), and we have
1
(b) =
2
(b) for any b = a
n
1
1
a
n
2
2
. . . a
n
k
n
with a
i
S or a

i
S for 1 i k. Since
1
and
2
are linear, they agree
on the linear span of such elements, which is dense by (i). Continuous maps
agreeing on a dense subset of their domain are equal, so
1
=
2
.
4.3.5 Lemma. 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.

Proof. Exercise.
We can now prove a generalisation of Proposition 4.1.12.
4.3.6 Corollary. If a is a normal element of a unital C*-algebra A then
r(a) = |a|.
Proof. Let B = C

(1, a). By Corollary 1.3.13, r(a) = r

A
(a) = r
B
(a). By the
Gelfand-Naimark theorem 4.3.1, if : B C((B)) is the Gelfand repre-
sentation of B then |a| = |(a)| = r
B
(a) = r(a).
4.3.7 Lemma. Suppose that a is a normal element of a unital C*-algebra, let
= (C

(1, a)) and let : C

(1, a) C() be the Gelfand representation

of C

(1, a). The map a = (a): C is a homeomorphism of onto (a).

Proof. By Lemma 3.1.4 and Proposition 4.1.9, every is a unital -
homomorphism C

(1, a) C. If a(
1
) = a(
2
) for some
1
,
2
then

1
(a) =
2
(a) and
1
(1) = 1 =
2
(1). Taking S = 1, a in Lemma 4.3.4(ii)
we see that
1
=
2
. Hence a is injective.
The map a is continuous by the denition of the topology on . More-
over, is compact by Theorem 3.1.6 and a() = (a) is Hausdor. By
Lemma 2.1.1, a is a homeomorphism onto (a).
4.3.8 Lemma. Suppose that X and Y are compact Hausdor topological
spaces and : X Y is a homeomorphism. The map
t
: C(Y ) C(X),
f f is an isometric unital -isomorphism.
Proof. Exercise.
38
4.3.9 Theorem (The continuous functional calculus). Let a be a normal
element of a unital C*-algebra A. There is a unique unital -homomorphism

a
: C((a)) A such that
a
(z) = a where z C((a)) is the function
z() = for (a). Moreover,
a
is an isometric -isomorphism onto
C

(1, a).
Proof. Suppose that
1
and
2
are two unital -homomorphisms C((a)) A
with
1
(z) =
2
(z) = a. Since z() = z() if and only if = , the
function z separates the points of (a), so C

(1, z) is a closed separating

unital -subalgebra of C((a)). By the Stone-Weierstrass theorem 4.2.3,
C

(1, z) = C((a)), so
1
=
2
by Lemma 4.3.4(ii). This proves the
uniqueness statement.
Let = (C

(1, a)), let : C

(1, a) C() be the Gelfand represen-

tation of C

(1, a) and let a = (a). By the Gelfand-Naimark theorem 4.3.1

and the last two lemmas, and (a)
t
: C((a)) C() are both isometric
unital -isomorphisms. Consider the map
a
: C((a)) C

(1, a) given by

a
=
1
(a)
t
:
C((a)) C

(1, a)

a
//
C((a))
C()
(a)
t

?
?
?
?
?
?
?
?
?
?
?
?
C()
C

(1, a)

1
??












Since it is the composition of two isometric -isomorphisms,
a
is an isometric
-isomorphism onto C

(1, a).
4.3.10 Denition. If a is a normal element of a unital C*-algebra A and
f C((a)) then we write f(a) for the element
a
(f) A.
4.3.11 Remark. If p =
0
+
1
z + +
n
z
n
is a complex polynomial
on (a), then
a
(p) is equal to the element p(a) dened in Denition 1.3.7,
because
a
is a unital homomorphism with
a
(z) = a. So this new notation
is consistent.
4.3.12 Remark. This notation is quite convenient. By way of example, we
use it to restate the fact that
a
is a -homomorphism. For any f, g C((a))
and C, we have
(f + g)(a) = f(a) + g(a), (f)(a) = f(a),
(fg)(a) = f(a)g(a) and f

(a) = f(a)

where the functions f + g, f, fg and f

are given by applying the usual

pointwise operations in the C*-algebra C((a)).
39
4.3.13 Theorem (Spectral mapping for the continuous functional calculus).
If a is a normal element of a unital C*-algebra A and f C((a)) then
(f(a)) = f((a)) = f(): (a).
Proof. Since
a
is a unital -isomorphism of C((a)) onto C

(1, a), we have

A
(f(a)) =
C

(1,a)
(f(a)) by Corollary 4.1.11
=
C

(1,a)
(
a
(f)) by the denition of f(a)
=
C((a))
(f) by Proposition 4.1.14(iv)
= f((a)) by Example 1.3.2(ii).
4.4 Positive elements of C*-algebras
Let us write R
+
for the non-negative real numbers.
4.4.1 Denition. Let A be a C*-algebra. If a A then we say that a is
positive and write a 0 if a is hermitian with (a) R
+
. The set of positive
elements of A will be denoted by A
+
= a A: a 0.
4.4.2 Example. In the case A = C(X) where X is a compact Hausdor
space, we have
C(X)
+
= f C(X): f(x) R
+
for x X
= f C(X): f = f

and |f t| t for some t R

+

= g

g : g C(X).
4.4.3 Corollary. Let A be a unital C*-algebra. For every a A
+
there is a
unique element b A
+
with b
2
= a.
Proof. Since (a) R
+
, the function f : (a) C, t

t is well-dened,
continuous and takes values in R
+
. Hence we can dene b = f(a), and since
f = f

we have b

= f

(a) = f(a) = b so b is hermitian. By Theorem 4.3.13,

(b) = f((a)) R
+
, so b A
+
. Since f
2
= z is the identity function
on (a), by Theorem 4.3.9 we have b
2
= f(a)
2
= (f
2
)(a) = z(a) = a. This
establishes the existence of b A
+
with b
2
= a.
To see that b is the unique element with these properties, suppose that
c A
+
with c
2
= a. Then ac = c
3
= ca, so B = C

(1, a, c) is a unital
abelian C*-algebra and b = f(a) C

(1, a) B. If (B) then

(b)
2
= (b
2
) = (a) = (c
2
) = (c)
2
;
since b, c A
+
we have (b), (c) 0, so (b) = (c). By Corollary 4.3.2,
b = c.
40
4.4.4 Denition. If A is a unital C*-algebra and a A
+
then we call the
element b A
+
with b
2
= a the positive square root of a, and write it as a
1/2
.
4.4.5 Lemma. Let A be a unital C*-algebra.
(i). If a is a hermitian element of A then a
2
0.
(ii). If a is a hermitian element of A then a 0 if and only if |a t| t
for some t R
+
.
(iii). If a, b A with a 0 and b 0 then a + b 0.
(iv). If a A with a 0 and a 0 then a = 0.
(v). If a A and a

a 0 then a = 0.
Proof. (i) Since (a) R by Corollary 4.1.10(i), the spectral mapping theo-
rem 1.3.8 gives (a
2
) =
2
: (a) R
+
, so a
2
0.
(ii) If a 0 then (a) [0, t] where t = r(a) R
+
. By Theorem 1.3.8
we have (a t) = (a) t [t, 0] and a t is hermitian, so we have
|a t| = r(a t) t by Proposition 4.1.12.
Conversely, if a = a

and |a t| t for some t R

+
then (a t)
[t, t], so (a) = (a t) + t [0, 2t] by Theorem 1.3.8. Hence a 0.
(iii) Clearly, a +b is hermitian, and by (ii) there exist s, t R
+
such that
|a s| s and |b t| t. By the triangle inequality,
|a + b (s + t)| |a s| +|b t| s + t,
so a + b 0 by (ii).
(iv) We have (a) R
+
and (a) = (a) R
+
, so (a) = 0. Hence
|a| = r(a) = 0 by Proposition 4.1.12, so a = 0.
(v) Using Proposition 4.1.7(iii), write a = h + ik where h and k are
hermitian elements of A. By Proposition 1.3.3(ii), aa

is also positive.
Now a

a + aa

= 2(h
2
+ k
2
), so a

a = 2(h
2
+ k
2
) aa

0 by (i) and (iii).

By (iv), a

a = 0, so |a|
2
= |a

a| = 0 and a = 0.
4.4.6 Theorem. If A is a unital C*-algebra then A
+
= a

a: a A.
Proof. If b A
+
then b = a
2
= a

a where a = b
1/2
, so A
+
a

a: a A.
It remains to show that if a A then b = a

a 0. By Proposi-
tion 4.1.7(i), b is hermitian. Consider the three functions z, z
+
, z

C((b))
given by
z() = , z
+
() =
_
if 0
0 if < 0
and z

() =
_
0 if 0
if < 0
41
Observe that z

((b)) R
+
, z = z
+
z

and z
+
z

= 0. Thus, writing b
+
=
z
+
(b) and b

= z

(b), we have b

0 by Theorem 4.3.13, and b = b

+
b

and b
+
b

= 0 by Theorem 4.3.9.
Let c = ab

. We have
c

c = b

ab

= b

(b
+
b

)b

= (b

)
3
0.
By Lemma 4.4.5(v), c = 0, so (b

)
3
= 0. Hence (b

) = 0; since b

is
hermitian, Proposition 4.1.12 shows that b

= 0, so b = b
+
0.
4.4.7 Denition. If a, b A then let us write a b if b a 0.
4.4.8 Remark. It follows from parts (iii) and (iv) of Lemma 4.4.5 that the
relation is a partial order on A.
4.4.9 Lemma. If A is a unital C*-algebra and a A
+
then a |a|1.
Proof. Exercise.
4.4.10 Corollary. Let A be a unital C*-algebra.
(i). If a, b A with a b then c

ac c

bc for any c A.
(ii). If a, b A then b

ab |a|b

b.
Proof. (i) Let d = (b a)
1/2
. Then c

bc c

ac = c

d
2
c = (dc)

(dc) 0 by
Theorem 4.4.6.
(ii) We have a

a |a|
2
1 by Lemma 4.4.9, so b

ab b

(|a|
2
1)b =
|a|
2
b

b by (i).
4.5 The GNS representation
If V is a vector space, recall that a positive semi-denite sesquilinear form
on V is a mapping (, ): V V C such that
(x, x) 0, (x + y, z) = (x, z) + (y, z) and (y, x) = (x, y)
for all x, y, z V and , C. If this mapping also satises
(x, x) = 0 = x = 0
then we say that it is a positive denite, and it is then an inner product on V
(see [FA4.1]).
You will probably have seen the next result proven for inner products,
but perhaps not for positive semi-denite sesquilinear forms. The proof is
basically the same, though.
42
4.5.1 Lemma (The Cauchy-Schwarz inequality). Let V be a vector space
and let (, ): V V C be a positive semi-denite sesquilinear form. Then
[(x, y)[
2
(x, x) (y, y) for all x, y V .
Proof. For C we have
0 (x y, x y) = [[
2
(x, x) (x, y) (y, x) + (y, y).
Taking = t(x, y) for t R gives
p(t) = t
2
[(x, y)[
2
(x, x) 2t[(x, y)[
2
+ (y, y) 0 for all t R.
Observe that p(t) is a polynomial in t. If p(t) is constant then (x, y) = 0
and we are done. Otherwise, [(x, y)[
2
,= 0 and since all non-constant linear
polynomials take negative values, p(t) must have degree 2.
Since p(t) 0 for all t R, its discriminant b
2
4ac is not positive.
Hence 4[(x, y)[
4
4[(x, y)[
2
(x, x)(y, y) 0, so [(x, y)[
2
(x, x)(y, y).
4.5.2 Denition. Let A be a unital C*-algebra. We say that a linear func-
tional : A C is positive if (a) 0 for all a A
+
.
4.5.3 Remark. If is a positive linear functional on A and a, b A with
a b (see Remark 4.4.8) then b a 0 so (b a) = (b) (a) 0, so
(a) (b). Thus positive linear functionals are order-preserving.
4.5.4 Lemma. Let A be a unital C*-algebra and let be a positive linear
functional on A.
(i). (a

) = (a) for all a A.

(ii). [(a)[ (1) |a| for all a A. Hence A

and || = (1).
Proof. (i) If h is a hermitian element of A then (h) [t, ) for some
t 0, so h +t1 0. Hence (h +t1) = (h) +t(1) 0. Since (1) 0, we
must have (h) R. Now if a A then a = h +ik for hermitian elements h
and k, so (a

) = (h ik) = (h) i(k) = (h) + i(k) = (a).

(ii) If a A then a

a 0 by Theorem 4.4.6, so a

a |a

a|1 = |a|
2
1 by
Lemma 4.4.9. Hence (a

a) (|a|
2
1) = (1)|a|
2
.
Consider the map (, ): AA C given by (a, b) = (b

a). Using (i) it

is easy to check that this is a positive semi-denite sesquilinear form, so by
the Cauchy-Schwarz inequality we have
[(a)[
2
= [(a, 1)[
2
(1, 1) (a, a) = (1) (a

a) (1)
2
|a|
2
.
Hence [(a)[ (1)|a|, and the result follows.
43
4.5.5 Lemma. Let A be a unital C*-algebra and let A

. Then is
positive if and only if || = (1).
Proof. If is positive then || = (1) by Lemma 4.5.4(ii). Conversely,
suppose that || = (1). Multiplying by a positive constant if necessary,
we may assume that || = (1) = 1. By Lemma 4.1.8, (a) is real whenever
a is hermitian. If a 0 and |a| 1 then |1 a| = r(1 a) 1. Hence
[1 (a)[ = [(1 a)[ 1.
Since (a) is real, this shows that (a) 0 so is positive.
4.5.6 Denition. Let A be a unital C*-algebra. A state on A is a positive
linear functional of norm 1. We write S(A) for the set of states of A. Thus
by the preceding lemma,
S(A) = A

: || = (1) = 1.
4.5.7 Remark. If A is an abelian unital C*-algebra then (A) S(A)
by Lemma 3.1.4. Thus states generalise the characters in the abelian case.
Moreover, if we turn S(A) into a topological space by giving it the subspace
topology from the weak* topology on A

, then S(A) = J
1
1
(1) A

1
is a
compact Hausdor space.
4.5.8 Lemma. Let A be a unital C*-algebra. If a is a hermitian element
of A then there is a state S(A) with [(a)[ = |a|.
Proof. Let B = C

(1, a), which is a unital abelian C*-algebra. By Corol-

lary 3.1.11(ii), there is
0
(B) with [
0
(a)[ = r(a), and r(a) = |a| by
Proposition 4.1.12. Moreover, we have
0
(1) = 1 = |
0
| by Lemma 3.1.4.
By the Hahn-Banach theorem [FA3.6], we can extend
0
to a functional
A

with || 1. Since (1) =

0
(1) = 1 we have || = (1) = 1 so
S(A), and |a| = [
0
(a)[ = [(a)[.
4.5.9 Lemma. Let A be a unital C*-algebra. For every S(A) there is a
Hilbert space H and a -homomorphism : A B(H) such that
(a

a) |(a)|
2
|a|
2
for all a A.
Proof. Consider the mapping (, ): A A C dened by
(a, b) = (b

a), a, b A.
It is easy to check that this is a positive semi-denite sesquilinear form.
Observe that if a A then a

a |a|
2
1 by Lemma 4.4.9, so
(a, a) = (a

a) |a|
2
(1) = |a|
2
.
44
Moreover, if a, b, c A then (ab, c) = (c

ab) = ((a

c)

b) = (b, a

c).
By the Cauchy-Schwarz inequality 4.5.1, for each a A we have
(a, a) = 0 (a, b) = 0 for all b A.
Hence if N = a A: (a, a) = 0 then
N = a A: (a, b) = 0 for all b A.
Using this expression, we can check that N is a left ideal of A. Indeed, if
a, b N then (a + b, c) = (a, c) + (b, c) = 0 and (a, c) = (a, c) = 0 and
(ca, d) = (a, c

d) = 0 for all c, d A and C, so a+b, a and ca are in N.

Consider the vector space V = A/N and let us write [a] = a +N V for
a A. We dene a mapping , ): V V C by
[a], [b]) = (a, b), a, b A.
This is well-dened since if [a
1
] = [a
2
] and [b
1
] = [b
2
] then a
2
a
1
N and
b
2
b
1
N, so (a
1
, b
1
) = (a
1
, b
1
) + (a
2
a
1
, b
1
) + (a
2
, b
2
b
1
) = (a
2
, b
2
). It
follows that , ) is a positive semi-denite sesquilinear form on V . In fact, it
is positive denite since if [a], [a]) = 0 then (a, a) = 0 so a N, i.e. [a] = 0.
Hence (V, , )) is an inner product space.
Let H be the completion of this inner product space. It is not hard to
show that , ) extends to an inner product on H turning H into a Hilbert
space.
Let L(V ) denote the set of linear maps V V , and let
0
: A L(V )
be given by
0
(a)[b] = [ab] for [b] V . This is well-dened since if [b
1
] = [b
2
]
then b
2
b
1
N, so if a A then a(b
2
b
1
) = ab
2
ab
1
N (since N is a left
ideal) and so [ab
1
] = [ab
2
]. Moreover, for a, b A we have b

ab |a|
2
b

b
by Corollary 4.4.10, so
|
0
(a)[b]|
2
= |[ab]|
2
= (ab, ab) = (b

ab) |a|
2
(b

b) = |a|
2
|[b]|
2
.
Hence |
0
(a)[b]| |a| |[b]|, so
0
(a) is a continuous linear map by [FA1.8.1],
and its operator norm satises |
0
(a)| |a|.
Thus
0
(a) has a unique extension to a map in B(H), written (a), and
|(a)| |a|. This denes a mapping : A B(H).
We claim that is a -homomorphism. Since V is dense in H and (a)
is continuous for all a A, it suces to check these properties on V . For
a, b, c A and C we have
(a + b)[c] = [(a + b)c] = [ac + bc] = (a)[c] + (b)[c] =
_
(a) + (b)
_
[c] and
(a)[c] = [ac] = (a)[c] so is linear,
(ab)[c] = [abc] = (a)[bc] = (a)(b)[c], so is a homomorphism, and
(a)[b], [c]) = (ab, c) = (b, a

c) = [b], (a

)[c]) so (a

) = (a)

.
45
Hence is a -homomorphism.
It remains to establish the inequality (a

a) |(a)|
2
|a|
2
. We have
already remarked that |(a)| |a|. Now |[1]|
2
= (1, 1) = (1) = 1, so
|(a)|
2
|(a)[1]|
2
= |[a]|
2
= (a, a) = (a

a).
We can now show that every C*-algebra is, up to isometric -isomorphism,
a subalgebra of B(H) for some Hilbert space H. To do so, we will piece to-
gether all of the representations constructed in Lemma 4.5.9. Since there are
a lot of these representations, we will need to take care of a few technicalities
rst.
Let I be an index set. If we are given
i
0 for each i I, then we will
say that (
i
)
iI
is summable if the set of nite sums
_

iJ

i
: J is a nite subset of I
_
is bounded above, and then we declare the value of

iI

i
to be supremum
of this set. If (
i
)
iI
is summable, we write

iI

i
< .
Suppose that t
i
R and

iI
[t
i
[ < . Writing J = j I : t
j
0
we can dene

iI
t
i
=

jJ
t
j

kI\J
(t
k
). Similarly, if
i
C and

iI
[
i
[ < then we can dene

iI

i
=

iI
Re(
i
) + i

iI
Im(
i
).
If H
i
: i I is a family of Hilbert spaces, then we can dene an inner
product space H as follows:
H =
_
(
i
)
iI
:
i
H
i
for each i I, and

iI
|
i
|
2
<
_
with pointwise vector space operations (
i
)
iI
+ (
i
)
iI
= (
i
+
i
)
iI
,
(
i
)
iI
= (
i
)
iI
and inner product
(
i
)
iI
, (
i
)
iI
) =

iI

i
,
i
).
It follows from the Cauchy-Schwarz inequality that

iI
[
i
,
i
)[ < if
(
i
)
iI
, (
i
)
iI
H, so the inner product is well-dened.
It is not too hard to show that H is then a complete inner product space,
i.e. H is a Hilbert space. We usually write
H =

iI
H
i
and call H the Hilbertian direct sum of the Hilbert spaces H
i
: i I.
46
Suppose that a
i
B(H
i
) for each i I and that sup
iI
|a
i
| < . We
may dene an operator a B(H) by a((
i
)
iI
) = (a(
i
))
iI
for (
i
)
iI
H.
Note that

iI
|a
i

i
|
2
(sup
iI
|a
i
|)

iI
|
i
|
2
for each x X, so a is a
bounded operator, with |a| sup
iI
|a
i
|. In fact, it is easy to show that
|a| = sup
iI
|a
i
| [exercise]. We will write a =

iI
a
i
.
4.5.10 Theorem (The Gelfand-Naimark-Segal theorem). If A is a unital
C*-algebra then A is isometrically -isomorphic to a subalgebra of B(H) for
some Hilbert space H.
Proof. Given S(A), let us write H

and

: A B(H

) for the Hilbert

space and the -homomorphism obtained from Lemma 4.5.9. Let
H =

S(A)
H

and dene (a) =

S(A)

(a) for a A.
It is easy to see that this denes a -homomorphism : A B(H). If a A
then since |(a)| = sup
S(A)
|

(a)|, we have
sup
S(A)
(a

a) |(a)|
2
|a|
2
by Lemma 4.5.9. However, a

a is hermitian so by Lemma 4.5.8,

sup
S(A)
(a

a) |a

a| = |a|
2
.
Putting these inequalities together shows that |(a)| = |a| for each a A,
so is an isometric -homomorphism. In particular, the range of is a closed
-subalgebra of B(H) which is -isomorphic to A.
47